ZhenYi/gitdataai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: 1.0

Browse Source
This commit is contained in:
zhenyi 2026-05-30 01:38:27 +08:00
parent 2b543f5e37
commit e1330451a5
2277 changed files with 22298 additions and 232612 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

88
.dockerignore
View File

@ -1,13 +1,79 @@
.git/ # Git
.idea/ .git
.vscode/ .gitignore
node_modules/
*.log
.env
.env.local
.env.*.local
# Exclude all target/ content, then selectively re-include release binaries # IDE
.idea
.vscode
*.swp
*.swo
*~
# Rust build artifacts
target/ target/
!target/release/ **/target/
!target/x86_64-unknown-linux-gnu/release/
# Node.js
node_modules/
npm-debug.log*
yarn-debug.log*
yarn-error.log*
pnpm-debug.log*
bun.lockb
# Build output
dist/
build/
# Environment and secrets
.env
.env.*
!.env.example
# Docker
docker/
docker-compose*.yml
.dockerignore
Dockerfile*
*.Dockerfile
# Documentation
*.md
LICENSE
doc/
# Test and CI
tests/
__tests__/
*.test.*
*.spec.*
.github/
.gitlab/
.circleci/
# OS files
.DS_Store
Thumbs.db
desktop.ini
# Logs
logs/
*.log
# Temporary files
tmp/
temp/
.tmp/
# Certificates (use secrets in production)
cert/
# Data directories
data/
# Agent configs
.agent/
.agents/
.claude/
CLAUDE.md
AGENTS.md

136
.env.example
View File

@ -1,136 +0,0 @@
# =============================================================================
# Required - 程序启动必须配置
# =============================================================================
# 数据库连接
APP_DATABASE_URL=postgresql://user:password@localhost:5432/dbname
APP_DATABASE_SCHEMA_SEARCH_PATH=public
# Redis（支持多节点，逗号分隔）
APP_REDIS_URL=redis://localhost:6379
# APP_REDIS_URLS=redis://localhost:6379,redis://localhost:6378
# AI 服务
APP_AI_BASIC_URL=https://api.openai.com/v1
APP_AI_API_KEY=sk-xxxxx
# Embedding + 向量检索
APP_EMBED_MODEL_BASE_URL=https://api.openai.com/v1
APP_EMBED_MODEL_API_KEY=sk-xxxxx
APP_EMBED_MODEL_NAME=text-embedding-3-small
APP_EMBED_MODEL_DIMENSIONS=1536
APP_QDRANT_URL=http://localhost:6333
# APP_QDRANT_API_KEY=
# SMTP 邮件
APP_SMTP_HOST=smtp.example.com
APP_SMTP_PORT=587
APP_SMTP_USERNAME=noreply@example.com
APP_SMTP_PASSWORD=xxxxx
APP_SMTP_FROM=noreply@example.com
APP_SMTP_TLS=true
APP_SMTP_TIMEOUT=30
# 文件存储
APP_AVATAR_PATH=/data/avatars
# Git 仓库存储根目录
APP_REPOS_ROOT=/data/repos
# =============================================================================
# Domain / URL（可选，有默认值）
# =============================================================================
APP_DOMAIN_URL=http://127.0.0.1
# APP_STATIC_DOMAIN=
# APP_MEDIA_DOMAIN=
# APP_GIT_HTTP_DOMAIN=
# =============================================================================
# Database Pool（可选，有默认值）
# =============================================================================
# APP_DATABASE_MAX_CONNECTIONS=10
# APP_DATABASE_MIN_CONNECTIONS=2
# APP_DATABASE_IDLE_TIMEOUT=60000 (milliseconds, default: 60s)
# APP_DATABASE_MAX_LIFETIME=300000 (milliseconds, default: 300s)
# APP_DATABASE_CONNECTION_TIMEOUT=5000 (milliseconds, default: 5s)
# APP_DATABASE_REPLICAS=
# APP_DATABASE_HEALTH_CHECK_INTERVAL=30
# APP_DATABASE_RETRY_ATTEMPTS=3
# APP_DATABASE_RETRY_DELAY=5
# =============================================================================
# Redis Pool（可选，有默认值）
# =============================================================================
# APP_REDIS_POOL_SIZE=10
# APP_REDIS_CONNECT_TIMEOUT=5
# APP_REDIS_ACQUIRE_TIMEOUT=5
# =============================================================================
# SSH（可选，有默认值）
# =============================================================================
# APP_SSH_DOMAIN=
# APP_SSH_PORT=22
# APP_SSH_SERVER_PRIVATE_KEY=
# APP_SSH_SERVER_PUBLIC_KEY=
# =============================================================================
# Logging（可选，有默认值）
# =============================================================================
# APP_LOG_LEVEL=info
# APP_LOG_FORMAT=json
# APP_LOG_FILE_ENABLED=false
# APP_LOG_FILE_PATH=./logs
# APP_LOG_FILE_ROTATION=daily
# APP_LOG_FILE_MAX_FILES=7
# APP_LOG_FILE_MAX_SIZE=104857600
# OpenTelemetry（可选，默认关闭）
# APP_OTEL_ENABLED=false
# APP_OTEL_ENDPOINT=http://localhost:5080/api/default/v1/traces
# APP_OTEL_SERVICE_NAME=
# APP_OTEL_SERVICE_VERSION=
# APP_OTEL_AUTHORIZATION=
# APP_OTEL_ORGANIZATION=
# =============================================================================
# NATS / Hook Pool（可选，有默认值）
# =============================================================================
# HOOK_POOL_MAX_CONCURRENT=（CPU 核数）
# HOOK_POOL_CPU_THRESHOLD=80.0
# HOOK_POOL_REDIS_LIST_PREFIX={hook}
# HOOK_POOL_REDIS_LOG_CHANNEL=hook:logs
# HOOK_POOL_REDIS_BLOCK_TIMEOUT=5
# HOOK_POOL_REDIS_MAX_RETRIES=3
# HOOK_POOL_WORKER_ID=（随机 UUID）
# =============================================================================
# Frontend (Vite) — 前端运行环境变量
# =============================================================================
# API 基础 URL（为空时使用 Vite dev 代理 /api -> localhost:8080）
# VITE_API_BASE_URL=http://localhost:8080
# 前端 WebSocket 连接地址（开发模式通过 Vite 代理）
VITE_WS_URL=ws://localhost:5080
# API URL（前端 API 调用，通过 Vite 代理时可为空）
VITE_API_URL=
# WebSocket 连接模式: "raw-ws" | "socketio"
VITE_WS_MODE=raw-ws
# =============================================================================
# Frontend: Grafana Faro (RUM) — 前端性能监控（可选）
# =============================================================================
# VITE_FARO_ENABLED=false
# VITE_FARO_URL=https://faro.example.com/collect
# VITE_FARO_API_KEY=
# VITE_FARO_APP_NAME=GitDataAIWeb
# VITE_FARO_APP_ENV=production
# VITE_FARO_APP_VERSION=0.0.1

83
.gitignore vendored
View File

@ -1,29 +1,62 @@
# Rust build artifacts
/target /target
node_modules **/target/
.claude
.zed # Rust IDE and tooling
.vscode .idea/
.idea .vscode/
*.swp
*.swo
*~
# Environment files
.env .env
.env.local .env.local
dist .env.*.local
deploy/secrets.yaml .env.production
.codex
.qwen # OS files
.opencode .DS_Store
.omc Thumbs.db
AGENT.md desktop.ini
ARCHITECTURE.md
.agents # Node.js
.agents.md
.next
node_modules/ node_modules/
coverage/ npm-debug.log*
.pnpm-store/ yarn-debug.log*
pnpm-lock.yaml yarn-error.log*
package-lock.json pnpm-debug.log*
yarn.lock
.gemini # Build output
.omg dist/
/.sqry build/
deploy/.server.yaml
# Logs
logs/
*.log
# Data and certificates
data/
cert/
# Docker
docker-compose.override.yml
.docker/
# Agent configs
.claude/
.codex/
.agent/
.agents/
CLAUDE.md
AGENTS.md
migrate.sh
# Temporary files
tmp/
temp/
.tmp/
# Backup files
*.bak
*.backup
*~

11
.mcp.json
View File

@ -1,11 +0,0 @@
{
"mcpServers": {
"shadcn": {
"command": "npx",
"args": [
"shadcn@latest",
"mcp"
]
}
}
}

7
.prettierignore
View File

@ -1,7 +0,0 @@
node_modules/
coverage/
.pnpm-store/
pnpm-lock.yaml
package-lock.json
pnpm-lock.yaml
yarn.lock

11
.prettierrc
View File

@ -1,11 +0,0 @@
{
"endOfLine": "lf",
"semi": false,
"singleQuote": false,
"tabWidth": 2,
"trailingComma": "es5",
"printWidth": 80,
"plugins": ["prettier-plugin-tailwindcss"],
"tailwindStylesheet": "src/index.css",
"tailwindFunctions": ["cn", "cva"]
}

7059
Cargo.lock generated
View File

File diff suppressed because it is too large Load Diff

345
Cargo.toml
View File

@ -1,192 +1,5 @@
[workspace]
members = [
# "libs/frontend",
"libs/models",
"libs/session",
"libs/git",
"libs/email",
"libs/queue",
"libs/room",
"libs/config",
"libs/service",
"libs/db",
"libs/api",
"libs/transport",
"libs/observability",
"libs/avatar",
"libs/agent",
"libs/migrate",
"libs/fctool",
"libs/gingress-proxy",
"apps/migrate",
"apps/app",
"apps/git-hook",
"apps/gitserver",
"apps/email",
"apps/static",
"apps/metrics",
"apps/gingress",
]
resolver = "3"
[workspace.dependencies]
models = { path = "libs/models" }
session = { path = "libs/session" }
git = { path = "libs/git" }
email = { path = "libs/email" }
queue = { path = "libs/queue" }
room = { path = "libs/room" }
config = { path = "libs/config" }
service = { path = "libs/service" }
db = { path = "libs/db" }
api = { path = "libs/api" }
agent = { path = "libs/agent" }
observability = { path = "libs/observability" }
avatar = { path = "libs/avatar" }
migrate = { path = "libs/migrate" }
fctool = { path = "libs/fctool" }
transport = { path = "libs/transport" }
metrics-aggregator = { path = "apps/metrics" }
gingress-proxy = { path = "libs/gingress-proxy" }
gingress = { path = "apps/gingress" }
sea-query = "1.0.0-rc.33"
actix-web = "4.13.0"
actix-files = "0.6.10"
actix-cors = "0.7.1"
actix-session = "0.11.0"
actix-ws = "0.4.0"
actix-multipart = "0.7.2"
actix-analytics = "1.2.1"
actix-jwt-session = "1.0.7"
actix-csrf = "0.8.0"
metrics = "0.24.5"
actix-rt = "2.11.0"
actix = "0.13"
async-stream = "0.3"
actix-service = "2.0.3"
actix-utils = "3.0.1"
redis = "1.1.0"
anyhow = "1.0.102"
derive_more = "2.1.1"
blake3 = "1.8.3"
argon2 = "0.5.3"
thiserror = "2.0.18"
password-hash = "0.6.0"
awc = "3.8.2"
bstr = "1.12.1"
captcha-rs = "0.5.0"
deadpool-redis = "0.23.0"
deadpool = "0.13.0"
dotenv = "0.15.0"
env_logger = "0.11.10"
brotli = "7.0"
flate2 = "1.1.9"
git2 = "0.20.4"
slog = "2.8.2"
git2-ext = "1.0.0"
git2-hooks = "0.7.0"
futures = "0.3.32"
futures-util = "0.3.32"
globset = "0.4.18"
hex = "0.4.3"
lettre = { version = "0.11.19", default-features = false, features = ["tokio1-rustls-tls", "smtp-transport", "builder", "pool"] }
mime = "0.3.17"
mime_guess2 = "2.3.1"
opentelemetry = "0.31.0"
opentelemetry-otlp = { version = "0.31.0", features = ["http-proto", "trace"] }
opentelemetry_sdk = { version = "0.31.0", features = ["rt-tokio"] }
opentelemetry-http = "0.31.0"
prost = "0.14.3"
prost-build = "0.14.3"
qdrant-client = "1.17.0"
prost-types = "0.14.3"
rand = "0.10.0"
russh = { version = "0.60.2", default-features = false, features = ["ring", "rsa"] }
hmac = { version = "0.13" }
hkdf = "0.13.0"
sha1_smol = "1.0.1"
rsa = { version = "0.9.7", package = "rsa" }
reqwest = { version = "0.13.2", default-features = false }
dotenvy = "0.15.7"
# aws-lc-sys requires NASM on Windows, so we use local filesystem storage instead of S3
# aws-sdk-s3 = "1.127.0"
sea-orm = "2.0.0-rc.37"
sea-orm-migration = "2.0.0-rc.37"
sha1 = "0.11"
sha2 = "0.11"
sysinfo = "0.39.1"
ssh-key = "0.7.0-rc.9"
tar = "0.4.45"
zip = "8.3.1"
tokenizer = "0.1.2"
tiktoken-rs = "0.11.0"
regex = "1.12.3"
jsonwebtoken = "10.3.0"
once_cell = "1.21.4"
async-trait = "0.1.89"
fs2 = "0.4.3"
image = "0.25.10"
tokio = "1.50.0"
tokio-util = "0.7.18"
tokio-stream = { version = "0.1.18", features = ["sync"] }
url = "2.5.8"
tower = "0.5"
num_cpus = "1.17.0"
ring = "0.17"
rustls = { version = "0.23", default-features = false, features = ["ring", "std", "tls12"] }
clap = "4.6.0"
time = "0.3.47"
chrono = "0.4.44"
tracing = "0.1.44"
tracing-subscriber = { version = "0.3.23", features = ["env-filter", "json", "tracing-log"] }
tracing-opentelemetry = "0.32.1"
tonic = "0.14.5"
tonic-build = "0.14.5"
uuid = "1.22.0"
hostname = "0.4"
utoipa = { version = "5.4.0", features = ["chrono", "uuid"] }
rust_decimal = "1.40.0"
walkdir = "2.5.0"
calamine = "0.26"
csv = "1.3"
lopdf = "0.34"
pulldown-cmark = "0.12"
quick-xml = { version = "0.37", features = ["serialize"] }
sqlparser = "0.55"
lazy_static = "1.5"
chacha20poly1305 = "0.10"
md5 = "0.7"
moka = "0.12.15"
dashmap = "7.0.0-rc2"
serde = "1.0.228"
serde_json = "1.0.149"
serde_yaml = "0.9.33"
serde_bytes = "0.11.19"
phf = "0.13.1"
phf_codegen = "0.13.1"
base64 = "0.22.1"
base64ct = "1"
p256 = { version = "0.13", features = ["ecdsa", "std"] }
# http version varies per-crate (pingora needs 1.x, actix needs 0.2)
hyper = "0.14"
tempfile = "3"
rig-core = { version = "0.36.0", default-features = false }
tokio-tungstenite = { version = "0.29.0", features = [] }
async-nats = { version = "0.48.0", features = [] }
kube = { version = "3.1.0", features = ["runtime", "derive"] }
k8s-openapi = { version = "0.27", features = ["v1_31"] }
pingora = { version = "0.8", features = ["proxy"] }
pingora-proxy = "0.8"
pingora-load-balancing = "0.8"
pingora-cache = "0.8"
rustls-pemfile = "2"
[workspace.package] [workspace.package]
version = "0.2.9" version = "1.0.0"
edition = "2024" edition = "2024"
authors = [] authors = []
description = "" description = ""
@ -198,6 +11,31 @@ keywords = []
categories = [] categories = []
documentation = "" documentation = ""
[workspace]
members = [
"app/email",
"app/gitdata",
"app/gitpod",
"app/gitsync",
"lib/ai",
"lib/api",
"lib/cache",
"lib/channel",
"lib/config",
"lib/db",
"lib/email",
"lib/git",
"lib/issues",
"lib/migrate",
"lib/model",
"lib/queue",
"lib/service",
"lib/session",
"lib/storage",
"lib/parsefile"
, "lib/socketio"]
resolver = "3"
[workspace.lints.rust] [workspace.lints.rust]
unsafe_code = "warn" unsafe_code = "warn"
@ -205,36 +43,105 @@ unsafe_code = "warn"
unwrap_used = "warn" unwrap_used = "warn"
expect_used = "warn" expect_used = "warn"
[profile.dev] [workspace.dependencies]
debug = 1 ai = { path = "lib/ai" }
incremental = true api = { path = "lib/api" }
codegen-units = 256 cache = { path = "lib/cache" }
channel = { path = "lib/channel" }
config = { path = "lib/config" }
db = { path = "lib/db" }
email = { path = "lib/email" }
git = { path = "lib/git" }
issues = { path = "lib/issues" }
migrate = { path = "lib/migrate" }
model = { path = "lib/model" }
queue = { path = "lib/queue" }
service = { path = "lib/service" }
session = { path = "lib/session" }
storage = { path = "lib/storage" }
parsefile = { path = "lib/parsefile"}
socketio = { path = "lib/socketio" }
[profile.release] leptos = "0.8.19"
lto = "thin" leptos_actix = "0.8.7"
codegen-units = 1 leptos_meta = "0.8.6"
strip = true leptos_router = "0.8.13"
opt-level = 3 server_fn = { version = "0.8.10", features = ["actix"] }
actix-http = "3.11"
actix-ws = "0.4.0"
urlencoding = "2.1"
serde_urlencoded = "0.7"
juniper = "0.17.1"
[profile.dev.package.num-bigint-dig] ractor = "0.15.13"
opt-level = 3 ractor_cluster = "0.15.13"
async-nats = "0.48.0"
petgraph = "0.8.3"
[package] async-openai = "0.40.0"
name = "workspace" rig-core = { version = "0.36.0", default-features = false, features = ["derive"] }
version.workspace = true schemars = "1.2.1"
edition.workspace = true tokio-stream = "0.1.18"
authors.workspace = true duct = "1.1.1"
description.workspace = true lettre = "0.11.22"
repository.workspace = true actix-web = "4"
readme.workspace = true jsonwebtoken = { version = "10.4.0", features = ["rust_crypto"] }
homepage.workspace = true futures-util = "0.3"
license.workspace = true futures = "0.3.32"
keywords.workspace = true moka = "0.12.15"
categories.workspace = true tokio = "1.52.3"
documentation.workspace = true redis = "1.2.1"
serde_json = "1.0.149"
[lib] indexmap = "2.14.0"
path = "lib.rs" sea-orm-migration = "2.0.0-rc.38"
crate-type = ["lib"] sea-orm = { version = "2.0.0-rc.38", features = ["sqlx-all","runtime-tokio","rust_decimal","uuid","chrono"]}
async-trait = "0.1.89"
aws-config = "1.8.16"
aws-sdk-s3 = "1.132.0"
rust_decimal = "1.42.0"
utoipa = "5.5.0"
dotenvy = "0.15.7"
anyhow = "1.0.102"
derive_more = "2.1.1"
serde = "1.0.228"
serde_yaml = "0.9.33"
comrak = "0.38"
sqlparser = "0.62.0"
qdrant-client = "1.18.0"
tiktoken-rs = "0.11.0"
tracing-subscriber = "0.3.23"
thiserror = "2.0.18"
uuid = "1.23.1"
git2 = "0.21.0"
gix = { version = "0.83.0", features = ["max-performance-safe", "serde", "merge", "blame", "revision", "blob-diff", "worktree-stream", "worktree-archive", "mailmap"] }
gix-archive = "0.32.0"
gix-worktree-stream = "0.32.0"
num_cpus = "1.17.0"
tracing = "0.1.44"
actix-service = "2.0.3"
actix-rt = "2.11.0"
actix-utils = "3.0.1"
toasty = "0.6.1"
chrono = "0.4.44"
argon2 = "0.5.3"
rand = "0.10.1"
rand_core = { version = "0.10.1", features = ["getrandom"] }
totp-rs = "5.7.1"
url = "2.5.7"
sha2 = "0.11.0"
base64 = "0.22"
tonic = "0.14.6"
tonic-build = "0.14.6"
prost = "0.14.3"
tonic-prost = "0.14.6"
dashmap = "6"
sqlx = "0.9.0"
russh = { version = "0.61.1", features = ["legacy-ed25519-pkcs8-parser"] }
hex = "0.4"
async-stream = "0.3"
tokio-util = "0.7"
password-hash = "0.6.1"
deadpool-redis = { version = "0.23", features = ["cluster"] }
reqwest = { version = "0.13", features = ["json", "rustls", "system-proxy"] }
hmac = "0.13"
mcpkit = "0.5"
miette = "7"

21
README.md
View File

@ -1,21 +0,0 @@
# React + TypeScript + Vite + shadcn/ui
This is a template for a new Vite project with React, TypeScript, and shadcn/ui.
## Adding components
To add components to your app, run the following command:
```bash
npx shadcn@latest add button
```
This will place the ui components in the `src/components` directory.
## Using components
To use the components in your app, import them as follows:
```tsx
import { Button } from "@/components/ui/button"
```

39
apps/app/Cargo.toml
View File

@ -1,39 +0,0 @@
[package]
name = "app"
version.workspace = true
edition.workspace = true
authors.workspace = true
description.workspace = true
repository.workspace = true
readme.workspace = true
homepage.workspace = true
license.workspace = true
keywords.workspace = true
categories.workspace = true
documentation.workspace = true
[dependencies]
tokio = { workspace = true, features = ["full"] }
uuid = { workspace = true }
service = { workspace = true }
observability = { workspace = true }
room = { workspace = true }
sha2 = { workspace = true }
hkdf = { workspace = true }
hmac = { workspace = true }
api = { workspace = true }
session = { workspace = true }
config = { workspace = true }
db = { workspace = true }
migrate = { workspace = true }
actix-web = { workspace = true }
actix-cors = { workspace = true }
futures = { workspace = true }
tracing = { workspace = true }
anyhow = { workspace = true }
clap = { workspace = true }
sea-orm = { workspace = true }
serde_json = { workspace = true }
chrono = { workspace = true }
[lints]
workspace = true

12
apps/app/src/args.rs
View File

@ -1,12 +0,0 @@
use clap::Parser;
#[derive(Parser, Debug)]
#[command(name = "app")]
#[command(version)]
pub struct ServerArgs {
#[arg(long, short)]
pub bind: Option<String>,
#[arg(long)]
pub workers: Option<usize>,
}

133
apps/app/src/logging.rs
View File

@ -1,133 +0,0 @@
//! Structured HTTP request logging middleware using tracing.
//!
//! Logs every incoming request with method, path, status code,
//! response time, client IP, authenticated user ID, and trace_id.
use actix_web::dev::{Service, ServiceRequest, ServiceResponse, Transform};
use futures::future::{LocalBoxFuture, Ready, ok};
use session::SessionExt;
use std::sync::Arc;
use std::task::{Context, Poll};
use std::time::Instant;
use uuid::Uuid;
/// Default log format: `{method} {path} {status} {duration_ms}ms`
pub struct RequestLogger {
trace_id_header: String,
}
impl RequestLogger {
pub fn new(trace_id_header: String) -> Self {
Self { trace_id_header }
}
}
impl<S, B> Transform<S, ServiceRequest> for RequestLogger
where
S: Service<ServiceRequest, Response = ServiceResponse<B>, Error = actix_web::Error> + 'static,
S::Future: 'static,
B: 'static,
{
type Response = ServiceResponse<B>;
type Error = actix_web::Error;
type Transform = RequestLoggerMiddleware<S>;
type InitError = ();
type Future = Ready<Result<Self::Transform, Self::InitError>>;
fn new_transform(&self, service: S) -> Self::Future {
ok(RequestLoggerMiddleware {
service: Arc::new(service),
trace_id_header: self.trace_id_header.clone(),
})
}
}
pub struct RequestLoggerMiddleware<S> {
service: Arc<S>,
trace_id_header: String,
}
impl<S> Clone for RequestLoggerMiddleware<S> {
fn clone(&self) -> Self {
Self {
service: self.service.clone(),
trace_id_header: self.trace_id_header.clone(),
}
}
}
impl<S, B> Service<ServiceRequest> for RequestLoggerMiddleware<S>
where
S: Service<ServiceRequest, Response = ServiceResponse<B>, Error = actix_web::Error> + 'static,
S::Future: 'static,
B: 'static,
{
type Response = ServiceResponse<B>;
type Error = actix_web::Error;
type Future = LocalBoxFuture<'static, Result<Self::Response, Self::Error>>;
fn poll_ready(&self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
self.service.poll_ready(cx)
}
fn call(&self, req: ServiceRequest) -> Self::Future {
let started = Instant::now();
let trace_id_header = self.trace_id_header.clone();
let method = req.method().to_string();
let path = req.path().to_string();
let query = req.query_string().to_string();
let remote = req
.connection_info()
.realip_remote_addr()
.map(|s| s.to_string())
.unwrap_or_else(|| "unknown".to_string());
let user_id: Option<Uuid> = req.get_session().user();
let trace_id = Uuid::now_v7().to_string();
let full_path = if query.is_empty() {
path.clone()
} else {
format!("{}?{}", path, query)
};
let service = self.service.clone();
Box::pin(async move {
let res = service.call(req).await?;
let elapsed = started.elapsed();
let status = res.status();
let status_code = status.as_u16();
let is_health = path == "/health";
if !is_health {
let user_id_str = user_id
.map(|u: Uuid| u.to_string())
.unwrap_or_else(|| "-".to_string());
let duration_ms = elapsed.as_millis() as u64;
let log_args = (
method = %method,
path = %full_path,
status = status_code,
duration_ms = duration_ms,
remote = %remote,
user_id = %user_id_str,
trace_id = %trace_id,
);
match status_code {
200..=299 => {
tracing::info!(log_args, "http_request");
}
400..=499 => {
tracing::warn!(log_args, "http_request");
}
_ => {
tracing::error!(log_args, "http_request");
}
}
}
Ok(res)
})
}
}

350
apps/app/src/main.rs
View File

@ -1,350 +0,0 @@
use actix_cors::Cors;
use actix_web::cookie::time::Duration;
use actix_web::dev::{Service, ServiceRequest, ServiceResponse};
use actix_web::{App, HttpResponse, HttpServer, cookie::Key, web};
use api::{robots, sidemap};
use clap::Parser;
use db::cache::AppCache;
use db::database::AppDatabase;
use futures::future::LocalBoxFuture;
use observability::{
HttpMetrics, HttpSnapshotGuard, MetricsMiddleware, TracingSpanMiddleware,
init_tracing_subscriber, install_recorder, prometheus_handler, push::MetricsPusher,
spawn_http_metrics_poller,
};
use sea_orm::ConnectionTrait;
use service::AppService;
use session::SessionMiddleware;
use session::config::{PersistentSession, SessionLifecycle, TtlExtensionPolicy};
use session::storage::RedisClusterSessionStore;
use std::sync::Arc;
use std::task::{Context, Poll};
use std::time::Instant;
mod args;
use args::ServerArgs;
use config::AppConfig;
use migrate::{Migrator, MigratorTrait};
#[derive(Clone)]
pub struct AppState {
pub db: AppDatabase,
pub cache: AppCache,
}
/// Custom middleware that logs requests except for noisy paths.
struct RequestLogger;
impl<S, B> actix_web::dev::Transform<S, ServiceRequest> for RequestLogger
where
S: Service<ServiceRequest, Response = ServiceResponse<B>, Error = actix_web::Error>,
S::Future: 'static,
B: 'static,
{
type Response = ServiceResponse<B>;
type Error = actix_web::Error;
type Transform = RequestLoggerService<S>;
type InitError = ();
type Future = futures::future::Ready<Result<Self::Transform, Self::InitError>>;
fn new_transform(&self, service: S) -> Self::Future {
futures::future::ok(RequestLoggerService {
service,
_marker: std::marker::PhantomData,
})
}
}
struct RequestLoggerService<S> {
service: S,
_marker: std::marker::PhantomData<fn(ServiceRequest)>,
}
impl<S, B> actix_web::dev::Service<ServiceRequest> for RequestLoggerService<S>
where
S: actix_web::dev::Service<
ServiceRequest,
Response = ServiceResponse<B>,
Error = actix_web::Error,
>,
S::Future: 'static,
B: 'static,
{
type Response = ServiceResponse<B>;
type Error = actix_web::Error;
type Future = LocalBoxFuture<'static, Result<Self::Response, Self::Error>>;
fn poll_ready(&self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
self.service.poll_ready(cx)
}
fn call(&self, req: ServiceRequest) -> Self::Future {
let path = req.path().to_string();
let method = req.method().to_string();
let should_log = !(path == "/health"
|| path == "/metrics"
|| path.starts_with("/ws")
|| path.starts_with("/assets"));
let start = Instant::now();
let fut = self.service.call(req);
Box::pin(async move {
let res = fut.await?;
if should_log {
tracing::info!(
target: "http_request",
method = %method,
path = %path,
status = res.status().as_u16(),
elapsed = ?start.elapsed(),
"{} {} {} {:?}",
method,
path,
res.status().as_u16(),
start.elapsed()
);
}
Ok(res)
})
}
}
fn build_session_key(cfg: &AppConfig) -> anyhow::Result<Key> {
if let Some(secret) = cfg.env.get("APP_SESSION_SECRET") {
if secret.len() < 32 {
tracing::warn!(
secret_len = secret.len(),
"APP_SESSION_SECRET is too short (<32 bytes), using generated key instead"
);
return Ok(Key::generate());
}
use hkdf::Hkdf;
use sha2::Sha256;
// HKDF-SHA256: standard key derivation with info string for domain separation
let hk = Hkdf::<Sha256>::new(Some(b"session-cookie-key"), secret.as_bytes());
let mut okm = [0u8; 64];
hk.expand(b"actix-session-signing-key", &mut okm)
.map_err(|e| anyhow::anyhow!("HKDF expand failed: {}", e))?;
return Ok(Key::from(&okm));
}
tracing::warn!(
"APP_SESSION_SECRET not set, using generated key (sessions invalidated on restart)"
);
Ok(Key::generate())
}
#[tokio::main]
async fn main() -> anyhow::Result<()> {
let cfg = AppConfig::load();
let log_level = cfg.log_level().unwrap_or_else(|_| "info".to_string());
let otel_enabled = cfg.otel_enabled().unwrap_or(false);
init_tracing_subscriber(&log_level, false);
tracing::info!(
app_name = %cfg.app_name().unwrap_or_default(),
app_version = %cfg.app_version().unwrap_or_default(),
"Starting application"
);
let db = AppDatabase::init(&cfg).await?;
tracing::info!("Database connected");
let redis_urls = cfg.redis_urls()?;
let store: RedisClusterSessionStore = RedisClusterSessionStore::new(redis_urls).await?;
tracing::info!("Redis connected");
let cache = AppCache::init(&cfg).await?;
tracing::info!("Cache initialized");
run_migrations(&db).await?;
let session_key = build_session_key(&cfg)?;
let args = ServerArgs::parse();
let service = AppService::new(cfg.clone()).await?;
tracing::info!("AppService initialized");
let _model_sync_handle = service.clone().start_sync_task();
// TODO: workspace module not yet wired — billing alert task pending
// let _billing_alert_handle = service.clone().start_billing_alert_task();
let (shutdown_tx, shutdown_rx) = tokio::sync::broadcast::channel::<()>(1);
let worker_service = service.clone();
let worker_handle =
tokio::spawn(async move { worker_service.start_room_workers(shutdown_rx).await });
let _otel_guard = if otel_enabled {
let endpoint = cfg
.otel_endpoint()
.unwrap_or_else(|_| "http://localhost:4317".to_string());
let service_name = cfg
.otel_service_name()
.unwrap_or_else(|_| "app".to_string());
let service_version = cfg
.otel_service_version()
.unwrap_or_else(|_| "0.1.0".to_string());
tracing::info!(endpoint = %endpoint, service = %service_name, "OTLP tracing enabled");
let guard =
observability::init_otlp(&endpoint, &service_name, &service_version, &log_level)
.map_err(|e| anyhow::anyhow!("OTLP init failed: {}", e))?;
guard
} else {
None
};
let prometheus_handle = install_recorder();
let prometheus_handle_data = web::Data::new(prometheus_handle.clone());
let http_metrics = std::sync::Arc::new(HttpMetrics::new());
let http_snapshot: HttpSnapshotGuard = std::sync::Arc::new(std::sync::RwLock::new(
observability::HttpMetricsSnapshot::default(),
));
let http_snapshot_for_poller = http_snapshot.clone();
spawn_http_metrics_poller(
http_metrics.clone(),
http_snapshot_for_poller,
std::time::Duration::from_secs(15),
);
let http_snapshot_data = web::Data::new(http_snapshot);
// Metrics pusher: periodically push all metrics to apps/metrics aggregator
if let Some(push_url) = std::env::var("METRICS_PUSH_URL").ok() {
let pusher = MetricsPusher::new(&push_url, "app");
pusher.spawn(
http_metrics.clone(),
Arc::new(prometheus_handle.clone()),
std::time::Duration::from_secs(15),
);
tracing::info!(push_url = %push_url, "Metrics pusher started (interval 15s)");
}
let bind_addr = args.bind.unwrap_or_else(|| "127.0.0.1:8080".to_string());
tracing::info!(bind_addr = %bind_addr, "Listening");
let http_metrics_server = http_metrics.clone();
let cors_origins: Vec<String> = cfg
.env
.get("CORS_ORIGINS")
.map(|s| {
s.split(',')
.map(|s| s.trim().to_string())
.filter(|s| !s.is_empty())
.collect()
})
.unwrap_or_else(|| vec!["http://localhost:5173".to_string()]);
let cookie_secure = cfg
.env
.get("APP_COOKIE_SECURE")
.map(|s| s != "false")
.unwrap_or(true);
tracing::info!(cookie_secure = cookie_secure, "Cookie secure mode");
HttpServer::new(move || {
let mut cors = Cors::default();
for origin in &cors_origins {
cors = cors.allowed_origin(origin);
}
let cors = cors
.allowed_methods(["GET", "POST", "PUT", "PATCH", "DELETE", "OPTIONS"])
.allowed_headers([
"Content-Type",
"Authorization",
"X-Requested-With",
"Accept",
"Origin",
])
.supports_credentials()
.max_age(3600);
let security_headers = actix_web::middleware::DefaultHeaders::new()
.add(("X-Content-Type-Options", "nosniff"))
.add(("X-Frame-Options", "DENY"))
.add(("Referrer-Policy", "strict-origin-when-cross-origin"));
let session_mw = SessionMiddleware::builder(store.clone(), session_key.clone())
.cookie_name("id".to_string())
.cookie_path("/".to_string())
.cookie_secure(cookie_secure)
.cookie_http_only(true)
.session_lifecycle(SessionLifecycle::PersistentSession(
PersistentSession::default()
.session_ttl(Duration::days(30))
.session_ttl_extension_policy(TtlExtensionPolicy::OnEveryRequest),
))
.build();
let metrics_mw = MetricsMiddleware::new(http_metrics_server.clone());
App::new()
.wrap(cors)
.wrap(security_headers)
.wrap(session_mw)
.wrap(RequestLogger)
.wrap(metrics_mw)
.wrap(TracingSpanMiddleware::new())
.app_data(web::Data::new(AppState {
db: db.clone(),
cache: cache.clone(),
}))
.app_data(web::Data::new(service.clone()))
.app_data(web::Data::new(cfg.clone()))
.app_data(web::Data::new(db.clone()))
.app_data(web::Data::new(cache.clone()))
.app_data(http_snapshot_data.clone())
.app_data(prometheus_handle_data.clone())
.route("/robots.txt", web::get().to(robots::robots))
.route("/sitemap.xml", web::get().to(sidemap::sitemap))
.service(
web::scope("/sidemap")
.route("", web::get().to(sidemap::sitemap))
.route("/static", web::get().to(sidemap::sitemap_static))
.route("/users", web::get().to(sidemap::sitemap_users))
.route("/projects", web::get().to(sidemap::sitemap_projects))
.route("/repos", web::get().to(sidemap::sitemap_repos)),
)
.route("/health", web::get().to(health_check))
.route("/metrics", web::get().to(prometheus_handler))
.configure(api::route::init_routes)
})
.bind(&bind_addr)?
.run()
.await?;
tracing::info!("Server stopped, shutting down room workers");
let _ = shutdown_tx.send(());
let _ = worker_handle.await;
tracing::info!("Room workers stopped");
Ok(())
}
async fn run_migrations(db: &AppDatabase) -> anyhow::Result<()> {
tracing::info!("Running database migrations...");
Migrator::up(db.writer(), None)
.await
.map_err(|e| anyhow::anyhow!("Migration failed: {:?}", e))?;
tracing::info!("Migrations completed");
Ok(())
}
async fn health_check(state: web::Data<AppState>) -> HttpResponse {
let db_ok = db_ping(&state.db).await;
let cache_ok = cache_ping(&state.cache).await;
let healthy = db_ok && cache_ok;
if healthy {
HttpResponse::Ok().json(serde_json::json!({
"status": "ok",
"db": "ok",
"cache": "ok",
}))
} else {
HttpResponse::ServiceUnavailable().json(serde_json::json!({
"status": "unhealthy",
"db": if db_ok { "ok" } else { "error" },
"cache": if cache_ok { "ok" } else { "error" },
}))
}
}
async fn db_ping(db: &AppDatabase) -> bool {
let writer_ok = db.writer().execute_unprepared("SELECT 1").await.is_ok();
let reader_ok = db.reader().execute_unprepared("SELECT 1").await.is_ok();
writer_ok && reader_ok
}
async fn cache_ping(cache: &AppCache) -> bool {
cache.conn().await.is_ok()
}

36
apps/email/Cargo.toml
View File

@ -1,36 +0,0 @@
[package]
name = "email-server"
version.workspace = true
edition.workspace = true
authors.workspace = true
description.workspace = true
repository.workspace = true
readme.workspace = true
homepage.workspace = true
license.workspace = true
keywords.workspace = true
categories.workspace = true
documentation.workspace = true
[[bin]]
name = "email-worker"
path = "src/main.rs"
[dependencies]
tokio = { workspace = true, features = ["full"] }
service = { workspace = true }
db = { workspace = true }
config = { workspace = true }
tracing = { workspace = true }
observability = { workspace = true }
anyhow = { workspace = true }
clap = { workspace = true, features = ["derive"] }
chrono = { workspace = true, features = ["serde"] }
hyper = { workspace = true }
serde_json = { workspace = true }
sea-orm = { workspace = true }
metrics = "0.22"
metrics-exporter-prometheus = "0.13"
[lints]
workspace = true

165
apps/email/src/main.rs
View File

@ -1,165 +0,0 @@
use clap::Parser;
use config::AppConfig;
use metrics::{Unit, describe_counter};
use metrics_exporter_prometheus::PrometheusHandle;
use observability::{HttpMetrics, init_tracing_subscriber, install_recorder, push::MetricsPusher};
use sea_orm::ConnectionTrait;
use service::AppService;
use std::sync::Arc;
#[derive(Parser, Debug)]
#[command(name = "email-worker")]
#[command(version)]
struct Args {
#[arg(long, default_value = "info")]
log_level: String,
}
async fn http_handler(
db: Arc<db::database::AppDatabase>,
cache: Arc<db::cache::AppCache>,
metrics: Arc<PrometheusHandle>,
req: hyper::Request<hyper::Body>,
) -> Result<hyper::Response<hyper::Body>, std::convert::Infallible> {
match req.uri().path() {
"/health" => {
let writer_ok = db.writer().execute_unprepared("SELECT 1").await.is_ok();
let reader_ok = db.reader().execute_unprepared("SELECT 1").await.is_ok();
let db_ok = writer_ok && reader_ok;
let cache_ok = cache.conn().await.is_ok();
let body = serde_json::json!({
"status": if db_ok && cache_ok { "ok" } else { "unhealthy" },
"db": if db_ok { "ok" } else { "error" },
"cache": if cache_ok { "ok" } else { "error" },
});
let status = if db_ok && cache_ok { 200 } else { 503 };
let body_bytes = match serde_json::to_string(&body) {
Ok(s) => hyper::Body::from(s),
Err(e) => {
return Ok(hyper::Response::builder()
.status(500)
.body(hyper::Body::from(format!("serialize error: {}", e)))
.expect("static response"));
}
};
Ok(hyper::Response::builder()
.status(status)
.header("content-type", "application/json")
.body(body_bytes)
.expect("static response"))
}
"/metrics" => {
let body = metrics.render();
Ok(hyper::Response::builder()
.status(200)
.header("content-type", "text/plain; version=0.0.4; charset=utf-8")
.body(hyper::Body::from(body))
.unwrap())
}
_ => Ok(hyper::Response::builder()
.status(404)
.body(hyper::Body::from("not found"))
.unwrap()),
}
}
#[tokio::main]
async fn main() -> anyhow::Result<()> {
let args = Args::parse();
let cfg = AppConfig::load();
init_tracing_subscriber(&args.log_level, false);
// Pre-register all email/queue metrics so they appear in /metrics even before first event.
describe_counter!(
"email_queued_total",
Unit::Count,
"Emails written to Redis stream"
);
describe_counter!(
"email_consumed_total",
Unit::Count,
"Emails consumed from queue"
);
describe_counter!(
"email_batch_size",
Unit::Count,
"Email consumer batch sizes accumulated"
);
describe_counter!(
"email_validation_skipped_total",
Unit::Count,
"Emails skipped due to invalid recipient"
);
describe_counter!(
"email_build_errors_total",
Unit::Count,
"Email message build failures"
);
describe_counter!(
"email_send_attempts_total",
Unit::Count,
"SMTP send attempts (including retries)"
);
describe_counter!("email_sent_total", Unit::Count, "Emails sent successfully");
describe_counter!(
"email_send_failures_total",
Unit::Count,
"Emails that failed after all retries"
);
let metrics_handle = Arc::new(install_recorder());
let http_metrics = Arc::new(HttpMetrics::new()); // Worker app — HTTP section will be empty
// Metrics pusher: periodically push all metrics to apps/metrics aggregator
if let Some(push_url) = std::env::var("METRICS_PUSH_URL").ok() {
let pusher = MetricsPusher::new(&push_url, "email");
pusher.spawn(
http_metrics.clone(),
metrics_handle.clone(),
std::time::Duration::from_secs(15),
);
tracing::info!(push_url = %push_url, "Metrics pusher started (interval 15s)");
}
tracing::info!("Starting email worker");
let service = AppService::new(cfg).await?;
let db = Arc::new(service.db.clone());
let cache = Arc::new(service.cache.clone());
let (shutdown_tx, shutdown_rx) = tokio::sync::broadcast::channel::<()>(1);
tokio::spawn(async move {
tokio::signal::ctrl_c().await.ok();
tracing::info!("shutting down email worker");
let _ = shutdown_tx.send(());
});
// Start health/metrics server on a dedicated port
let health_db = db.clone();
let health_cache = cache.clone();
let health_metrics = metrics_handle.clone();
let health_addr: std::net::SocketAddr = ([0, 0, 0, 0], 8084).into();
let health_service = hyper::service::make_service_fn(move |_| {
let db = health_db.clone();
let cache = health_cache.clone();
let metrics = health_metrics.clone();
let service = hyper::service::service_fn(move |req| {
http_handler(db.clone(), cache.clone(), metrics.clone(), req)
});
async move { Ok::<_, std::convert::Infallible>(service) }
});
let health_server = hyper::Server::bind(&health_addr).serve(health_service);
tracing::info!(port = 8084, "health/metrics server started");
tokio::spawn(async move {
if let Err(e) = health_server.await {
tracing::error!("health check server error: {}", e);
}
});
service.start_email_workers(shutdown_rx).await?;
tracing::info!("email worker stopped");
Ok(())
}

35
apps/git-hook/Cargo.toml
View File

@ -1,35 +0,0 @@
[package]
name = "git-hook"
version.workspace = true
edition.workspace = true
authors.workspace = true
description.workspace = true
repository.workspace = true
readme.workspace = true
homepage.workspace = true
license.workspace = true
keywords.workspace = true
categories.workspace = true
documentation.workspace = true
[dependencies]
tokio = { workspace = true, features = ["full"] }
git = { workspace = true }
observability = { workspace = true }
db = { workspace = true }
config = { workspace = true }
tracing = { workspace = true }
tracing-subscriber = { workspace = true, features = ["json"] }
anyhow = { workspace = true }
clap = { workspace = true, features = ["derive"] }
tokio-util = { workspace = true }
hyper = { workspace = true }
serde_json = { workspace = true }
sea-orm = { workspace = true }
metrics = "0.22"
metrics-exporter-prometheus = "0.13"
chrono = { workspace = true, features = ["serde"] }
reqwest = { workspace = true }
agent = { workspace = true }
models = { workspace = true }
async-trait = { workspace = true }

10
apps/git-hook/src/args.rs
View File

@ -1,10 +0,0 @@
use clap::Parser;
#[derive(Parser, Debug)]
#[command(name = "git-hook")]
#[command(version)]
pub struct HookArgs {
/// Worker ID for this instance. Defaults to the HOOK_POOL_WORKER_ID env var or a generated UUID.
#[arg(long)]
pub worker_id: Option<String>,
}

256
apps/git-hook/src/main.rs
View File

@ -1,256 +0,0 @@
use clap::Parser;
use config::AppConfig;
use db::cache::AppCache;
use db::database::AppDatabase;
use git::hook::HookService;
use git::hook::embed::TagEmbedder;
use metrics::{Unit, describe_counter};
use metrics_exporter_prometheus::PrometheusHandle;
use observability::{HttpMetrics, init_tracing_subscriber, install_recorder, push::MetricsPusher};
use sea_orm::ConnectionTrait;
use std::sync::Arc;
use tokio::signal;
mod args;
use args::HookArgs;
/// Initialize EmbedService from config (graceful degradation).
async fn init_embed_service(
cfg: &AppConfig,
db: &AppDatabase,
) -> Result<agent::embed::EmbedService, Box<dyn std::error::Error + Send + Sync>> {
let client = agent::new_embed_client(cfg).await?;
let model_name = cfg
.get_embed_model_name()
.unwrap_or_else(|_| "text-embedding-3-small".into());
let dimensions = cfg.get_embed_model_dimensions().unwrap_or(1536);
let svc = agent::embed::EmbedService::new(client, db.writer().clone(), model_name, dimensions);
let _ = svc.ensure_collections().await;
tracing::info!("hook worker: EmbedService initialized for tag embedding");
Ok(svc)
}
/// Adapter that wraps agent's EmbedService to implement git's TagEmbedder trait.
struct EmbedServiceAdapter(agent::embed::EmbedService);
#[async_trait::async_trait]
impl TagEmbedder for EmbedServiceAdapter {
async fn embed_tags_batch(
&self,
tags: Vec<models::TagEmbedInput>,
) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
// Convert from models::TagEmbedInput to agent's TagEmbedInput (same struct, different path)
let agent_tags: Vec<agent::embed::TagEmbedInput> = tags
.into_iter()
.map(|t| agent::embed::TagEmbedInput {
repo_id: t.repo_id,
repo_name: t.repo_name,
project_id: t.project_id,
name: t.name,
description: t.description,
})
.collect();
self.0
.embed_tags_batch(agent_tags)
.await
.map_err(|e| Box::new(e) as Box<dyn std::error::Error + Send + Sync>)
}
}
async fn http_handler(
db: Arc<AppDatabase>,
cache: Arc<AppCache>,
metrics: Arc<PrometheusHandle>,
req: hyper::Request<hyper::Body>,
) -> Result<hyper::Response<hyper::Body>, std::convert::Infallible> {
match req.uri().path() {
"/health" => {
let writer_ok = db.writer().execute_unprepared("SELECT 1").await.is_ok();
let reader_ok = db.reader().execute_unprepared("SELECT 1").await.is_ok();
let db_ok = writer_ok && reader_ok;
let cache_ok = cache.conn().await.is_ok();
let body = serde_json::json!({
"status": if db_ok && cache_ok { "ok" } else { "unhealthy" },
"db": if db_ok { "ok" } else { "error" },
"cache": if cache_ok { "ok" } else { "error" },
});
let status = if db_ok && cache_ok { 200 } else { 503 };
let body_bytes = match serde_json::to_string(&body) {
Ok(s) => hyper::Body::from(s),
Err(e) => {
return Ok(hyper::Response::builder()
.status(500)
.body(hyper::Body::from(format!("serialize error: {}", e)))
.expect("static response"));
}
};
Ok(hyper::Response::builder()
.status(status)
.header("content-type", "application/json")
.body(body_bytes)
.expect("static response"))
}
"/metrics" => {
let body = metrics.render();
Ok(hyper::Response::builder()
.status(200)
.header("content-type", "text/plain; version=0.0.4; charset=utf-8")
.body(hyper::Body::from(body))
.expect("static response"))
}
_ => Ok(hyper::Response::builder()
.status(404)
.body(hyper::Body::from("not found"))
.expect("static response")),
}
}
#[tokio::main]
async fn main() -> anyhow::Result<()> {
let cfg = AppConfig::load();
let log_level = cfg.log_level().unwrap_or_else(|_| "info".to_string());
init_tracing_subscriber(&log_level, false);
// Pre-register all hook metrics so they appear in /metrics even before first increment.
describe_counter!("hook_tasks_total", Unit::Count, "Total hook tasks dequeued");
describe_counter!(
"hook_tasks_success_total",
Unit::Count,
"Hook tasks completed successfully"
);
describe_counter!(
"hook_tasks_failed_total",
Unit::Count,
"Hook tasks that failed"
);
describe_counter!(
"hook_tasks_locked_total",
Unit::Count,
"Hook tasks re-queued due to repo lock"
);
describe_counter!(
"hook_tasks_retried_total",
Unit::Count,
"Hook tasks that entered retry"
);
describe_counter!(
"hook_tasks_exhausted_total",
Unit::Count,
"Hook tasks that exhausted retries"
);
describe_counter!(
"hook_sync_branches_changed_total",
Unit::Count,
"Branches changed during sync"
);
describe_counter!(
"hook_sync_tags_changed_total",
Unit::Count,
"Tags changed during sync"
);
let metrics_handle = Arc::new(install_recorder());
let http_metrics = Arc::new(HttpMetrics::new()); // Worker app — HTTP section will be empty
// Metrics pusher: periodically push all metrics to apps/metrics aggregator
if let Some(push_url) = std::env::var("METRICS_PUSH_URL").ok() {
let pusher = MetricsPusher::new(&push_url, "git-hook");
pusher.spawn(
http_metrics.clone(),
metrics_handle.clone(),
std::time::Duration::from_secs(15),
);
tracing::info!(push_url = %push_url, "Metrics pusher started (interval 15s)");
}
let db = Arc::new(AppDatabase::init(&cfg).await?);
tracing::info!("database connected");
// 4. Connect to Redis cache (also provides the cluster pool for hook queue)
let cache = Arc::new(AppCache::init(&cfg).await?);
tracing::info!("cache connected");
// 5. Parse CLI args
let _args = HookArgs::parse();
tracing::info!("git-hook worker starting");
// 6. Build and start git hook service
let mut hooks = HookService::new(
(*db).clone(),
(*cache).clone(),
cache.redis_pool().clone(),
cfg.clone(),
);
// Optionally initialize tag embedding
if let Ok(embed_svc) = init_embed_service(&cfg, &db).await {
let adapter = EmbedServiceAdapter(embed_svc);
hooks = hooks.with_tag_embedder(Arc::new(adapter));
}
let cancel = hooks.start_worker().await;
let cancel_signal = cancel.clone();
// 7. Start health/metrics server on a dedicated port
let health_db = db.clone();
let health_cache = cache.clone();
let health_metrics = metrics_handle.clone();
let health_addr: std::net::SocketAddr = ([0, 0, 0, 0], 8083).into();
let health_service = hyper::service::make_service_fn(move |_| {
let db = health_db.clone();
let cache = health_cache.clone();
let metrics = health_metrics.clone();
let service = hyper::service::service_fn(move |req| {
http_handler(db.clone(), cache.clone(), metrics.clone(), req)
});
async move { Ok::<_, std::convert::Infallible>(service) }
});
let health_server = hyper::Server::bind(&health_addr).serve(health_service);
tracing::info!(port = 8083, "health/metrics server started");
tokio::spawn(async move {
if let Err(e) = health_server.await {
tracing::error!("health check server error: {}", e);
}
});
// Spawn signal handler that cancels on SIGINT/SIGTERM
tokio::spawn(async move {
let ctrl_c = async {
signal::ctrl_c()
.await
.expect("failed to install CTRL+C handler");
};
#[cfg(unix)]
let term = async {
use tokio::signal::unix::{SignalKind, signal};
let mut sig =
signal(SignalKind::terminate()).expect("failed to install SIGTERM handler");
sig.recv().await;
};
#[cfg(not(unix))]
let term = std::future::pending::<()>();
tokio::select! {
_ = ctrl_c => {
tracing::info!("received SIGINT, initiating shutdown");
}
_ = term => {
tracing::info!("received SIGTERM, initiating shutdown");
}
}
cancel_signal.cancel();
});
// Wait until the worker is cancelled (by signal handler or otherwise)
cancel.cancelled().await;
tracing::info!("git-hook worker stopped");
Ok(())
}

31
apps/gitserver/Cargo.toml
View File

@ -1,31 +0,0 @@
[package]
name = "gitserver"
version.workspace = true
edition.workspace = true
authors.workspace = true
description.workspace = true
repository.workspace = true
readme.workspace = true
homepage.workspace = true
license.workspace = true
keywords.workspace = true
categories.workspace = true
documentation.workspace = true
[[bin]]
name = "gitserver"
path = "src/main.rs"
[dependencies]
tokio = { workspace = true, features = ["full"] }
git = { workspace = true }
observability = { workspace = true }
tracing = { workspace = true }
db = { workspace = true }
config = { workspace = true }
anyhow = { workspace = true }
clap = { workspace = true, features = ["derive"] }
chrono = { workspace = true, features = ["serde"] }
[lints]
workspace = true

59
apps/gitserver/src/main.rs
View File

@ -1,59 +0,0 @@
use clap::Parser;
use config::AppConfig;
use observability::{HttpMetrics, init_tracing_subscriber, install_recorder, push::MetricsPusher};
use std::sync::Arc;
#[derive(Parser, Debug)]
#[command(name = "gitserver")]
#[command(version)]
struct Args {
#[arg(long, default_value = "info")]
log_level: String,
}
#[tokio::main]
async fn main() -> anyhow::Result<()> {
let args = Args::parse();
let cfg = AppConfig::load();
init_tracing_subscriber(&args.log_level, false);
let prometheus_handle = Arc::new(install_recorder());
let http_metrics = Arc::new(HttpMetrics::new());
// Metrics pusher: periodically push all metrics to apps/metrics aggregator
if let Some(push_url) = std::env::var("METRICS_PUSH_URL").ok() {
let pusher = MetricsPusher::new(&push_url, "gitserver");
pusher.spawn(
http_metrics.clone(),
prometheus_handle.clone(),
std::time::Duration::from_secs(15),
);
tracing::info!(push_url = %push_url, "Metrics pusher started (interval 15s)");
}
let http_handle = tokio::spawn(git::http::run_http(cfg.clone()));
let ssh_handle = tokio::spawn(git::ssh::run_ssh(cfg));
tokio::select! {
result = http_handle => {
match result {
Ok(Ok(())) => tracing::info!("HTTP server stopped"),
Ok(Err(e)) => tracing::error!("HTTP server error: {}", e),
Err(e) => tracing::error!("HTTP server task panicked: {}", e),
}
}
result = ssh_handle => {
match result {
Ok(Ok(())) => tracing::info!("SSH server stopped"),
Ok(Err(e)) => tracing::error!("SSH server error: {}", e),
Err(e) => tracing::error!("SSH server task panicked: {}", e),
}
}
_ = tokio::signal::ctrl_c() => {
tracing::info!("received shutdown signal");
}
}
tracing::info!("shutting down");
Ok(())
}

58
apps/metrics/Cargo.toml
View File

@ -1,58 +0,0 @@
[package]
name = "metrics-aggregator"
version.workspace = true
edition.workspace = true
authors.workspace = true
description = "Unified observability aggregator: scrapes metrics, forwards traces, collects logs"
repository.workspace = true
readme.workspace = true
homepage.workspace = true
license.workspace = true
keywords.workspace = true
categories.workspace = true
documentation.workspace = true
[[bin]]
name = "metrics-aggregator"
path = "src/main.rs"
[dependencies]
tokio = { workspace = true, features = ["full"] }
config = { workspace = true }
tracing = { workspace = true }
tracing-subscriber = { workspace = true, features = ["env-filter", "json"] }
observability = { workspace = true }
anyhow = { workspace = true }
clap = { workspace = true, features = ["derive", "env"] }
serde_json = { workspace = true }
chrono = { workspace = true, features = ["serde"] }
serde = { workspace = true, features = ["derive"] }
# HTTP server
actix-web = "4.13.0"
actix-rt = "2.11.0"
# HTTP client for scraping (uses awc = actix-web client, no extra TLS deps)
awc = { workspace = true }
# HTTP client for Loki (reqwest is Send+Sync, unlike awc::Client)
reqwest = { workspace = true, features = ["json"] }
# Metrics
metrics = { workspace = true }
metrics-exporter-prometheus = { version = "0.18", default-features = false, features = ["http-listener", "tokio"] }
# Observability
opentelemetry = { workspace = true }
opentelemetry_sdk = { workspace = true }
opentelemetry-otlp = { version = "0.31.0", default-features = false, features = ["http-proto", "tokio", "trace", "tonic"] }
tracing-opentelemetry = "0.32.1"
tokio-util = { workspace = true }
tokio-stream = { workspace = true }
futures = { workspace = true }
url = { workspace = true }
tower = { workspace = true }
[lints]
workspace = true

35
apps/metrics/src/args.rs
View File

@ -1,35 +0,0 @@
use clap::Parser;
#[derive(Parser, Debug)]
#[command(name = "metrics-aggregator")]
#[command(version)]
pub struct Args {
#[arg(long, default_value = "9090", env = "METRICS_AGGREGATOR_PORT")]
pub port: u16,
#[arg(long, env = "OTEL_EXPORTER_OTLP_ENDPOINT")]
pub otel_endpoint: Option<String>,
#[arg(long, env = "LOKI_URL")]
pub loki_url: Option<String>,
#[arg(long, default_value = "15", env = "SCRAPE_INTERVAL_SECS")]
pub scrape_interval_secs: u64,
/// JSON file with scrape targets.
#[arg(long, env = "SCRAPE_TARGETS_FILE")]
pub targets_file: Option<String>,
#[arg(long, default_value = "info", env = "LOG_LEVEL")]
pub log_level: String,
/// Comma-separated list of app names to scrape.
#[arg(long, env = "SCRAPE_APPS")]
pub scrape_apps: Option<String>,
#[arg(long)]
pub no_otel: bool,
#[arg(long)]
pub no_loki: bool,
}

40
apps/metrics/src/hotreload.rs
View File

@ -1,40 +0,0 @@
use std::sync::Arc;
use tokio::sync::RwLock;
use crate::target::{ScrapeTarget, load_targets_from_file};
pub async fn watch_targets_file(
path: String,
targets: Arc<RwLock<Vec<ScrapeTarget>>>,
mut shutdown: tokio::sync::broadcast::Receiver<()>,
) {
let mtime_path = path;
let mut last_mtime: Option<std::time::SystemTime> = None;
loop {
tokio::select! {
_ = shutdown.recv() => break,
_ = tokio::time::sleep(std::time::Duration::from_secs(10)) => {
let metadata = match tokio::fs::metadata(&mtime_path).await {
Ok(m) => m,
Err(_) => continue,
};
let current_mtime = metadata.modified().ok();
if current_mtime != last_mtime {
last_mtime = current_mtime;
match load_targets_from_file(&mtime_path).await {
Ok(new_targets) => {
let mut guard = targets.write().await;
*guard = new_targets;
tracing::info!(path = %mtime_path, "targets file reloaded");
}
Err(e) => {
tracing::warn!(error = %e, "failed to reload targets file");
}
}
}
}
}
}
}

70
apps/metrics/src/k8s_discovery.rs
View File

@ -1,70 +0,0 @@
use std::time::Duration;
use awc::Client;
use crate::target::ScrapeTarget;
pub async fn k8s_pod_discovery() -> Option<Vec<ScrapeTarget>> {
let pod_namespace = std::env::var("POD_NAMESPACE").ok()?;
let token_path = "/var/run/secrets/kubernetes.io/serviceaccount/token";
let token = tokio::fs::read_to_string(token_path).await.ok()?;
let client = Client::builder()
.timeout(Duration::from_secs(5))
.add_default_header((
awc::http::header::AUTHORIZATION.as_str(),
format!("Bearer {}", token),
))
.finish();
let api_url = format!(
"https://kubernetes.default.svc/api/v1/namespaces/{}/pods",
pod_namespace
);
let mut response = client.get(api_url).send().await.ok()?;
let body_bytes = response.body().await.ok()?;
let pod_list: serde_json::Value = serde_json::from_slice(&body_bytes).ok()?;
let targets: Vec<ScrapeTarget> = pod_list["items"]
.as_array()?
.iter()
.filter_map(|pod| {
let name = pod["metadata"]["name"].as_str()?.to_string();
let phase = pod["status"]["phase"].as_str()?;
if phase != "Running" {
return None;
}
let pod_ip = pod["status"]["podIP"].as_str()?;
let annotations = pod["metadata"]["annotations"].as_object()?;
let port: u16 = annotations
.get("metrics.port")
.and_then(|v| v.as_str())
.and_then(|s| s.parse().ok())
.unwrap_or(8080);
let path = annotations
.get("metrics.path")
.and_then(|v| v.as_str())
.unwrap_or("/metrics");
let labels = pod["metadata"]["labels"]
.as_object()
.map(|m| {
m.iter()
.filter_map(|(k, v)| v.as_str().map(|s| (k.clone(), s.to_string())))
.collect()
})
.unwrap_or_default();
Some(ScrapeTarget {
name,
addr: format!("{}:{}", pod_ip, port),
metrics_path: path.to_string(),
labels,
})
})
.collect();
Some(targets)
}

70
apps/metrics/src/loki.rs
View File

@ -1,70 +0,0 @@
use chrono::{DateTime, Utc};
use reqwest::Client;
use serde::Serialize;
use std::collections::HashMap;
#[derive(Clone)]
pub struct LokiForwarder {
url: String,
client: Client,
labels: HashMap<String, String>,
}
impl LokiForwarder {
pub fn new(url: String) -> Self {
Self {
url,
client: Client::builder()
.timeout(std::time::Duration::from_secs(5))
.build()
.expect("valid reqwest client"),
labels: HashMap::new(),
}
}
pub async fn push(&self, log_entries: Vec<LokiEntry>) -> anyhow::Result<()> {
if log_entries.is_empty() {
return Ok(());
}
let streams: Vec<LokiStream> = vec![LokiStream {
stream: self.labels.clone(),
values: log_entries
.into_iter()
.map(|e| (format!("{}", e.timestamp), e.line))
.collect(),
}];
let payload = LokiPayload { streams };
let resp = self
.client
.post(&self.url)
.header("Content-Type", "application/json")
.json(&payload)
.send()
.await;
match resp {
Ok(r) if r.status().is_success() => Ok(()),
Ok(r) => anyhow::bail!("Loki push failed: {}", r.status()),
Err(e) => anyhow::bail!("Loki push error: {}", e),
}
}
}
#[derive(Serialize)]
struct LokiPayload {
streams: Vec<LokiStream>,
}
#[derive(Serialize)]
struct LokiStream {
stream: HashMap<String, String>,
values: Vec<(String, String)>,
}
pub struct LokiEntry {
pub timestamp: DateTime<Utc>,
pub line: String,
}

633
apps/metrics/src/main.rs
View File

@ -1,633 +0,0 @@
//! Unified observability aggregator for in-cluster deployment.
//!
//! Collects metrics from all app pods via Prometheus scrape, forwards traces
//! to OTLP endpoint, and streams logs from all pods to Loki-compatible backend.
//!
//! Usage:
//! METRICS_AGGREGATOR_PORT=9090 \
//! OTEL_EXPORTER_OTLP_ENDPOINT=http://otel-collector:4317 \
//! LOKI_URL=http://loki:3100/loki/api/v1/push \
//! SCRAPE_INTERVAL_SECS=15 \
//! SCRAPE_TARGETS_FILE=/etc/metrics/targets.json \
//! metrics-aggregator
mod args;
mod hotreload;
mod k8s_discovery;
mod loki;
mod metrics;
mod otel;
mod scrape;
mod stats_store;
mod target;
use serde::Deserialize;
use std::collections::HashMap;
use std::fmt::Write as _;
use std::net::SocketAddr;
use std::sync::Arc;
use std::time::Duration;
use actix_web::{HttpResponse, HttpServer, web};
use clap::Parser;
use loki::{LokiEntry, LokiForwarder};
use metrics::AggMetrics;
use observability::{init_tracing_subscriber, install_recorder, instance_id};
use otel::OtelGuard;
use scrape::{HttpClient, ScrapeResult};
use stats_store::StatsStore;
use target::ScrapeTarget;
use tokio::io::AsyncBufReadExt;
use tokio::sync::{RwLock, broadcast};
use tokio::time::interval;
type MetricsStore = Arc<RwLock<HashMap<String, Vec<scrape::PromMetric>>>>;
// StatsStore is defined in stats_store.rs — per-app aggregated data.
#[actix_web::main]
async fn main() -> std::io::Result<()> {
let args = args::Args::parse();
init_tracing_subscriber(&args.log_level, false);
let instance = instance_id();
tracing::info!(
instance = %instance,
port = args.port,
scrape_interval = args.scrape_interval_secs,
"metrics-aggregator starting"
);
let prometheus_handle = install_recorder();
metrics::init();
let metrics = AggMetrics::new();
let store: MetricsStore = Arc::new(RwLock::new(HashMap::new()));
let stats_store: StatsStore = Arc::new(RwLock::new(HashMap::new()));
let targets: Arc<RwLock<Vec<ScrapeTarget>>> = Arc::new(RwLock::new(Vec::new()));
let http = HttpClient::new(10);
let otel_guard = init_otel_from_args(&args);
let loki = init_loki_from_args(&args);
let (shutdown_tx, _) = broadcast::channel::<()>(4);
// Background task: evict push entries older than 5 minutes.
let stats_store_for_evict = stats_store.clone();
let mut evict_shutdown = shutdown_tx.subscribe();
tokio::spawn(async move {
let mut ticker = interval(Duration::from_secs(30));
loop {
tokio::select! {
_ = evict_shutdown.recv() => break,
_ = ticker.tick() => {
let cutoff = chrono::Utc::now().timestamp() - 300;
let mut guard = stats_store_for_evict.write().await;
guard.retain(|_, entry| entry.last_seen >= cutoff);
}
}
}
});
if let Some(path) = &args.targets_file {
match target::load_targets_from_file(path).await {
Ok(initial_targets) => {
let mut guard = targets.write().await;
*guard = initial_targets;
tracing::info!(count = guard.len(), "loaded initial targets from file");
}
Err(e) => {
tracing::warn!(error = %e, "failed to load targets file");
}
}
let tw =
hotreload::watch_targets_file(path.clone(), targets.clone(), shutdown_tx.subscribe());
tokio::spawn(tw);
} else if std::env::var("KUBERNETES_SERVICE_HOST").is_ok() {
if let Some(k8s_targets) = k8s_discovery::k8s_pod_discovery().await {
let mut guard = targets.write().await;
*guard = k8s_targets.clone();
tracing::info!(count = guard.len(), "discovered K8s pods as targets");
}
}
let scrape_filter = args
.scrape_apps
.as_ref()
.map(|s| s.split(',').map(|p| p.trim().to_string()).collect());
let scrape_targets = targets.clone();
let scrape_store = store.clone();
let scrape_metrics = metrics.clone();
let scrape_http = http.clone();
let loki_clone = loki.clone();
let shutdown_tx_clone = shutdown_tx.clone();
let scrape_interval = args.scrape_interval_secs;
let scrape_filter_clone = scrape_filter.clone();
tokio::task::spawn_local(async move {
scrape_loop(
scrape_targets,
scrape_store,
scrape_metrics,
scrape_http,
scrape_interval,
scrape_filter_clone,
loki_clone,
shutdown_tx_clone.subscribe(),
)
.await;
});
let log_shutdown = shutdown_tx.subscribe();
let log_loki = loki.clone();
tokio::task::spawn_local(async move {
log_collector(log_loki, log_shutdown).await;
});
let bind_addr: SocketAddr = ([0, 0, 0, 0], args.port).into();
tracing::info!(addr = %bind_addr, "HTTP server starting");
let app_targets = targets.clone();
let app_store = store.clone();
let app_handle = prometheus_handle.clone();
let loki_for_push: Option<Arc<LokiForwarder>> = loki.map(Arc::new);
let app_stats = stats_store.clone();
let server = HttpServer::new(move || {
let targets = app_targets.clone();
let store = app_store.clone();
let handle = app_handle.clone();
let stats_store = app_stats.clone();
let loki_for_push: Option<Arc<LokiForwarder>> = loki_for_push.clone();
actix_web::App::new()
.app_data(web::Data::new(targets))
.app_data(web::Data::new(store))
.app_data(web::Data::new(handle))
.app_data(web::Data::new(stats_store))
.app_data(web::Data::new(loki_for_push))
.route("/metrics", web::get().to(handle_metrics))
.route("/api/v1/metrics", web::get().to(handle_metrics))
.route("/api/v1/push", web::post().to(handle_push))
.route("/api/v1/dashboard", web::get().to(handle_dashboard))
.route("/api/v1/stats", web::get().to(handle_stats))
.route("/health", web::get().to(handle_health))
.route("/api/v1/health", web::get().to(handle_health))
.route("/api/v1/targets", web::get().to(handle_targets))
})
.bind(&bind_addr)?
.run();
let server_handle = server.handle();
tokio::spawn(server);
tokio::signal::ctrl_c().await.ok();
tracing::info!("received Ctrl+C, shutting down");
let _ = shutdown_tx.send(());
server_handle.stop(true).await;
if let Some(guard) = otel_guard {
guard.shutdown().await;
}
tracing::info!("metrics-aggregator stopped");
Ok(())
}
fn init_otel_from_args(args: &args::Args) -> Option<OtelGuard> {
if args.no_otel {
return None;
}
let endpoint = args
.otel_endpoint
.clone()
.or_else(|| std::env::var("OTEL_EXPORTER_OTLP_ENDPOINT").ok())?;
match otel::init_otel(&endpoint, "metrics-aggregator") {
Ok(guard) => {
tracing::info!(endpoint = %endpoint, "OTLP tracing enabled");
Some(guard)
}
Err(e) => {
tracing::warn!(error = %e, "OTLP init failed, continuing without traces");
None
}
}
}
fn init_loki_from_args(args: &args::Args) -> Option<LokiForwarder> {
if args.no_loki {
return None;
}
let url = args
.loki_url
.clone()
.or_else(|| std::env::var("LOKI_URL").ok())?;
tracing::info!("Loki log forwarding enabled");
Some(LokiForwarder::new(url))
}
async fn handle_metrics(
store: web::Data<MetricsStore>,
stats_store: web::Data<StatsStore>,
handle: web::Data<observability::PrometheusHandle>,
) -> HttpResponse {
let extra = vec![("aggregator_instance".to_string(), "default".to_string())];
let scraped = render_aggregated_metrics(store, extra.clone()).await;
let pushed = render_pushed_metrics(stats_store).await;
let combined = format!("{}{}{}", handle.render(), scraped, pushed);
HttpResponse::Ok()
.content_type("text/plain; version=0.0.4; charset=utf-8")
.body(combined)
}
async fn handle_health() -> HttpResponse {
HttpResponse::Ok()
.content_type("application/json")
.body(r#"{"status":"ok"}"#)
}
async fn handle_targets(targets: web::Data<Arc<RwLock<Vec<ScrapeTarget>>>>) -> HttpResponse {
let guard = targets.read().await;
let json = serde_json::to_string(&*guard).unwrap_or_default();
HttpResponse::Ok()
.content_type("application/json")
.body(json)
}
// ── Push endpoint payload ────────────────────────────────────────────────────
#[derive(Debug, Deserialize)]
#[serde(rename_all = "camelCase")]
struct PushPayload {
app: String,
#[serde(default)]
instance: String,
timestamp: i64,
#[serde(default)]
http: Option<observability::push::HttpPayload>,
#[serde(default)]
system: Option<observability::push::SystemPayload>,
#[serde(default)]
business: HashMap<String, f64>,
#[serde(default)]
token_usage: Option<observability::push::TokenUsagePayload>,
#[serde(default)]
tasks: Option<observability::push::TaskStatsPayload>,
#[serde(default)]
latency: HashMap<String, observability::push::LatencySnapshot>,
#[serde(default)]
logs: Vec<observability::push::LogEntry>,
}
async fn handle_push(
stats_store: web::Data<StatsStore>,
loki: web::Data<Option<Arc<LokiForwarder>>>,
payload: web::Json<PushPayload>,
) -> HttpResponse {
let app = payload.app.clone();
stats_store::merge_push_payload(
&stats_store,
&app,
&payload.instance,
payload.timestamp,
payload.http.as_ref(),
payload.system.as_ref(),
&payload.business,
payload.token_usage.as_ref(),
payload.tasks.as_ref(),
&payload.latency,
&payload.logs,
)
.await;
// Forward logs to Loki if configured
if !payload.logs.is_empty() {
if let Some(loki_fwd) = loki.as_ref() {
let entries: Vec<LokiEntry> = payload
.logs
.iter()
.map(|l| LokiEntry {
timestamp: chrono::DateTime::from_timestamp(l.timestamp, 0)
.unwrap_or_else(chrono::Utc::now),
line: format!("[{}] {}", l.level.to_lowercase(), l.message),
})
.collect();
if let Err(e) = loki_fwd.push(entries).await {
tracing::warn!(error = %e, "loki push on /push failed");
}
}
}
HttpResponse::Ok().body("ok")
}
async fn scrape_loop(
targets: Arc<RwLock<Vec<ScrapeTarget>>>,
store: MetricsStore,
metrics: AggMetrics,
http: HttpClient,
interval_secs: u64,
scrape_apps_filter: Option<Vec<String>>,
_loki: Option<LokiForwarder>,
mut shutdown: broadcast::Receiver<()>,
) {
let mut ticker = interval(Duration::from_secs(interval_secs));
loop {
tokio::select! {
_ = shutdown.recv() => break,
_ = ticker.tick() => {
let targets_snapshot = targets.read().await.clone();
let count = targets_snapshot.len() as u64;
metrics.targets_total.set(count as f64);
let mut healthy_count = 0u64;
for target in &targets_snapshot {
if let Some(ref filter) = scrape_apps_filter {
if !filter.contains(&target.name) {
continue;
}
}
metrics.scrape_total.increment(1);
match http.scrape(target).await {
ScrapeResult::Success(body, duration_ms) => {
metrics.scrape_success.increment(1);
metrics.scrape_duration.record(duration_ms);
let parsed = scrape::parse_prometheus(&body);
update_store(store.clone(), &target.name, parsed).await;
healthy_count += 1;
}
ScrapeResult::Timeout => {
metrics.scrape_failures.increment(1);
metrics.scrape_errors_timeout.increment(1);
tracing::warn!(target = %target.name, "scrape timeout");
}
ScrapeResult::ConnectionError(e) => {
metrics.scrape_failures.increment(1);
metrics.scrape_errors_connection.increment(1);
tracing::warn!(target = %target.name, error = %e, "scrape connection error");
}
ScrapeResult::HttpError(status) => {
metrics.scrape_failures.increment(1);
tracing::warn!(target = %target.name, status = status, "scrape HTTP error");
}
}
}
metrics.targets_healthy.set(healthy_count as f64);
}
}
}
}
async fn update_store(store: MetricsStore, target_name: &str, metrics: Vec<scrape::PromMetric>) {
let mut guard = store.write().await;
guard.insert(target_name.to_string(), metrics);
}
async fn render_aggregated_metrics(
store: web::Data<MetricsStore>,
extra_group_labels: Vec<(String, String)>,
) -> String {
let guard = store.read().await;
let mut output = String::new();
for (target_name, metrics) in guard.iter() {
for metric in metrics {
let mut labels = metric.labels.clone();
labels.insert(
"aggregated_by".to_string(),
"metrics-aggregator".to_string(),
);
labels.insert("source_target".to_string(), target_name.clone());
for (k, v) in &extra_group_labels {
labels.insert(k.clone(), v.clone());
}
let label_str = if labels.is_empty() {
String::new()
} else {
let pairs: Vec<String> = labels
.iter()
.map(|(k, v)| {
format!(
r#"{}="{}""#,
k,
v.replace('\\', "\\\\").replace('"', "\\\"")
)
})
.collect();
format!("{{{}}}", pairs.join(","))
};
let _ = writeln!(&mut output, "{}{} {}", metric.name, label_str, metric.value);
}
}
output
}
async fn render_pushed_metrics(stats_store: web::Data<StatsStore>) -> String {
let guard = stats_store.read().await;
let mut output = String::new();
for (app_name, entry) in guard.iter() {
let labels = [
format!(r#"app="{}""#, app_name),
"aggregated_by".to_string(),
"metrics-aggregator".to_string(),
"push_source=true".to_string(),
];
let label_str = format!("{{{}}}", labels.join(","));
let h = &entry;
let _ = writeln!(
&mut output,
"push_http_requests_total{} {}",
label_str, h.requests_total
);
let _ = writeln!(
&mut output,
"push_http_request_duration_ms_total{} {}",
label_str, h.request_duration_ms_total
);
let _ = writeln!(
&mut output,
"push_http_requests_2xx{} {}",
label_str, h.requests_2xx
);
let _ = writeln!(
&mut output,
"push_http_requests_4xx{} {}",
label_str, h.requests_4xx
);
let _ = writeln!(
&mut output,
"push_http_requests_5xx{} {}",
label_str, h.requests_5xx
);
for (endpoint, &count) in &h.endpoints {
let sanitized = endpoint.replace([' ', '/'], "_").to_lowercase();
let ep_labels = format!(
r#"app="{}",endpoint="{}",aggregated_by="metrics-aggregator",push_source="true""#,
app_name, sanitized
);
let _ = writeln!(
&mut output,
"push_http_endpoint_requests_total{{{}}} {}",
ep_labels, count
);
}
// System metrics in Prometheus format
let sys_labels = format!(r#"app="{}",aggregated_by="metrics-aggregator""#, app_name);
let _ = writeln!(
&mut output,
"system_cpu_usage_percent{{{}}} {}",
sys_labels, h.cpu_usage_percent
);
let _ = writeln!(
&mut output,
"system_memory_used_mb{{{}}} {}",
sys_labels, h.memory_used_mb
);
let _ = writeln!(
&mut output,
"system_memory_total_mb{{{}}} {}",
sys_labels, h.memory_total_mb
);
let _ = writeln!(
&mut output,
"system_uptime_secs{{{}}} {}",
sys_labels, h.uptime_secs
);
// Business counters
for (counter_name, value) in &h.business {
let biz_labels = format!(r#"app="{}",aggregated_by="metrics-aggregator""#, app_name);
let _ = writeln!(&mut output, "{}{{{}}} {}", counter_name, biz_labels, value);
}
// Token usage
let ai_labels = format!(r#"app="{}",aggregated_by="metrics-aggregator""#, app_name);
let _ = writeln!(
&mut output,
"ai_input_tokens_total{{{}}} {}",
ai_labels, h.ai_input_tokens_total
);
let _ = writeln!(
&mut output,
"ai_output_tokens_total{{{}}} {}",
ai_labels, h.ai_output_tokens_total
);
let _ = writeln!(
&mut output,
"ai_calls_total{{{}}} {}",
ai_labels, h.ai_calls_total
);
// Latency per endpoint
for (endpoint, lat) in &h.latency {
let lat_labels = format!(
r#"app="{}",endpoint="{}",aggregated_by="metrics-aggregator""#,
app_name, endpoint
);
let _ = writeln!(
&mut output,
"latency_p99_ms{{{}}} {}",
lat_labels, lat.p99_ms
);
let _ = writeln!(
&mut output,
"latency_p90_ms{{{}}} {}",
lat_labels, lat.p90_ms
);
let _ = writeln!(
&mut output,
"latency_p50_ms{{{}}} {}",
lat_labels, lat.p50_ms
);
let _ = writeln!(
&mut output,
"latency_max_ms{{{}}} {}",
lat_labels, lat.max_ms
);
}
}
output
}
// ── JSON API handlers ────────────────────────────────────────────────────────
async fn handle_dashboard(stats_store: web::Data<StatsStore>) -> HttpResponse {
let dashboard = stats_store::build_dashboard(&stats_store).await;
let json = serde_json::to_string(&dashboard).unwrap_or_default();
HttpResponse::Ok()
.content_type("application/json")
.body(json)
}
async fn handle_stats(stats_store: web::Data<StatsStore>) -> HttpResponse {
// Returns per-app stats as JSON
let guard = stats_store.read().await;
let json = serde_json::to_string(&*guard).unwrap_or_default();
HttpResponse::Ok()
.content_type("application/json")
.body(json)
}
async fn log_collector(loki: Option<LokiForwarder>, mut shutdown: broadcast::Receiver<()>) {
let stdin = tokio::io::stdin();
let mut reader = tokio::io::BufReader::new(stdin);
let mut interval_tick = interval(Duration::from_secs(1));
let mut batch: Vec<LokiEntry> = Vec::with_capacity(100);
let mut line_buf = String::new();
loop {
tokio::select! {
_ = shutdown.recv() => break,
_ = interval_tick.tick() => {
if !batch.is_empty() {
if let Some(ref loki) = loki {
if let Err(e) = loki.push(std::mem::take(&mut batch)).await {
tracing::warn!(error = %e, "Loki push failed");
}
}
}
}
_ = async { line_buf.clear(); reader.read_line(&mut line_buf).await.ok() } => {
if !line_buf.is_empty() {
let line = line_buf.trim_end().to_string();
if !line.is_empty() {
batch.push(LokiEntry {
timestamp: chrono::Utc::now(),
line,
});
if batch.len() >= 100 {
if let Some(ref loki) = loki {
if let Err(e) = loki.push(std::mem::take(&mut batch)).await {
tracing::warn!(error = %e, "Loki push failed");
}
}
}
}
}
}
}
}
}

101
apps/metrics/src/metrics.rs
View File

@ -1,101 +0,0 @@
use metrics::{
Counter, Gauge, Histogram, Unit, describe_counter, describe_gauge, describe_histogram,
};
pub fn init() {
describe_gauge!(
"aggregator_targets_total",
Unit::Count,
"Total number of scrape targets known to the aggregator"
);
describe_gauge!(
"aggregator_targets_healthy",
Unit::Count,
"Number of scrape targets that responded last scrape"
);
describe_counter!(
"aggregator_scrape_total",
Unit::Count,
"Total number of scrape attempts"
);
describe_counter!(
"aggregator_scrape_success",
Unit::Count,
"Successful scrapes"
);
describe_counter!(
"aggregator_scrape_failures",
Unit::Count,
"Failed scrape attempts"
);
describe_counter!(
"aggregator_scrape_errors_parse",
Unit::Count,
"Scrape failures due to parse errors"
);
describe_counter!(
"aggregator_scrape_errors_timeout",
Unit::Count,
"Scrape failures due to timeout"
);
describe_counter!(
"aggregator_scrape_errors_connection",
Unit::Count,
"Scrape failures due to connection errors"
);
describe_counter!(
"aggregator_targets_discovered",
Unit::Count,
"Total targets discovered"
);
describe_counter!(
"aggregator_targets_lost",
Unit::Count,
"Total targets that disappeared"
);
describe_histogram!(
"aggregator_scrape_duration_ms",
Unit::Milliseconds,
"Scrape duration in milliseconds"
);
}
#[derive(Clone)]
#[allow(dead_code)]
pub struct AggMetrics {
pub targets_total: Gauge,
pub targets_healthy: Gauge,
pub scrape_total: Counter,
pub scrape_success: Counter,
pub scrape_failures: Counter,
pub scrape_errors_parse: Counter,
pub scrape_errors_timeout: Counter,
pub scrape_errors_connection: Counter,
pub targets_discovered: Counter,
pub targets_lost: Counter,
pub scrape_duration: Histogram,
}
impl Default for AggMetrics {
fn default() -> Self {
Self {
targets_total: metrics::gauge!("aggregator_targets_total"),
targets_healthy: metrics::gauge!("aggregator_targets_healthy"),
scrape_total: metrics::counter!("aggregator_scrape_total"),
scrape_success: metrics::counter!("aggregator_scrape_success"),
scrape_failures: metrics::counter!("aggregator_scrape_failures"),
scrape_errors_parse: metrics::counter!("aggregator_scrape_errors_parse"),
scrape_errors_timeout: metrics::counter!("aggregator_scrape_errors_timeout"),
scrape_errors_connection: metrics::counter!("aggregator_scrape_errors_connection"),
targets_discovered: metrics::counter!("aggregator_targets_discovered"),
targets_lost: metrics::counter!("aggregator_targets_lost"),
scrape_duration: metrics::histogram!("aggregator_scrape_duration_ms"),
}
}
}
impl AggMetrics {
pub fn new() -> Self {
Self::default()
}
}

42
apps/metrics/src/otel.rs
View File

@ -1,42 +0,0 @@
use anyhow::Context;
use opentelemetry::trace::TracerProvider;
use opentelemetry_otlp::{SpanExporter, WithExportConfig};
use opentelemetry_sdk::trace as sdktrace;
use tracing_opentelemetry::layer;
use tracing_subscriber::prelude::*;
pub struct OtelGuard {
provider: sdktrace::SdkTracerProvider,
}
impl OtelGuard {
pub async fn shutdown(self) {
if let Err(e) = self.provider.shutdown() {
tracing::warn!(error = %e, "OTLP shutdown error");
}
}
}
pub fn init_otel(endpoint: &str, service_name: &str) -> anyhow::Result<OtelGuard> {
let exporter = SpanExporter::builder()
.with_http()
.with_endpoint(endpoint)
.build()
.context("build OTLP exporter")?;
let tracer_provider = sdktrace::SdkTracerProvider::builder()
.with_batch_exporter(exporter)
.build();
let tracer = tracer_provider.tracer(service_name.to_string());
let otel_layer = layer().with_tracer(tracer);
tracing_subscriber::registry()
.with(otel_layer)
.try_init()
.context("install OTLP tracing subscriber")?;
Ok(OtelGuard {
provider: tracer_provider,
})
}

135
apps/metrics/src/scrape.rs
View File

@ -1,135 +0,0 @@
use awc::Client;
use std::collections::HashMap;
use crate::target::ScrapeTarget;
#[derive(Clone)]
pub struct HttpClient {
client: Client,
}
impl HttpClient {
pub fn new(timeout_secs: u64) -> Self {
let client = Client::builder()
.timeout(std::time::Duration::from_secs(timeout_secs))
.finish();
Self { client }
}
pub async fn scrape(&self, target: &ScrapeTarget) -> ScrapeResult {
let start = std::time::Instant::now();
let url = target.url();
let mut resp = match self.client.get(url).send().await {
Ok(resp) => resp,
Err(e) => {
let msg = e.to_string();
if msg.contains("timeout") || msg.contains("TimedOut") || msg.contains("timed out")
{
return ScrapeResult::Timeout;
}
return ScrapeResult::ConnectionError(msg);
}
};
if !resp.status().is_success() {
return ScrapeResult::HttpError(resp.status().as_u16());
}
let body = match resp.body().await {
Ok(bytes) => String::from_utf8_lossy(&bytes).into_owned(),
Err(e) => return ScrapeResult::ConnectionError(e.to_string()),
};
let scrape_ms = start.elapsed().as_millis() as f64;
ScrapeResult::Success(body, scrape_ms)
}
}
pub enum ScrapeResult {
Success(String, f64),
Timeout,
ConnectionError(String),
HttpError(u16),
}
#[derive(Clone, Debug)]
pub struct PromMetric {
pub name: String,
pub value: f64,
pub labels: HashMap<String, String>,
}
pub fn parse_prometheus(body: &str) -> Vec<PromMetric> {
let mut metrics = Vec::new();
for line in body.lines() {
let line = line.trim();
if line.is_empty() || line.starts_with('#') {
continue;
}
let (name_and_labels, value_str) = match line.find(' ') {
Some(pos) => (&line[..pos], &line[pos + 1..]),
None => continue,
};
let value: f64 = match value_str
.split_whitespace()
.next()
.and_then(|v| v.parse().ok())
{
Some(v) => v,
None => continue,
};
let (metric_name, labels) = if let Some(brace) = name_and_labels.find('{') {
let name = &name_and_labels[..brace];
let label_str = &name_and_labels[brace + 1..name_and_labels.len() - 1];
let labels = parse_labels(label_str);
(name.to_string(), labels)
} else {
(name_and_labels.to_string(), HashMap::new())
};
metrics.push(PromMetric {
name: metric_name,
value,
labels,
});
}
metrics
}
pub fn parse_labels(s: &str) -> HashMap<String, String> {
let mut labels = HashMap::new();
let mut remaining = s;
while !remaining.is_empty() {
if let Some(eq) = remaining.find('=') {
let key = remaining[..eq].trim().to_string();
remaining = &remaining[eq + 1..];
let (value, rest) = if remaining.starts_with('"') {
let end = remaining[1..]
.find('"')
.map(|p| p + 1)
.unwrap_or(remaining.len());
(&remaining[1..end], &remaining[end + 1..])
} else if remaining.starts_with('\'') {
let end = remaining[1..]
.find('\'')
.map(|p| p + 1)
.unwrap_or(remaining.len());
(&remaining[1..end], &remaining[end + 1..])
} else {
let end = remaining
.find(|c: char| !c.is_alphanumeric() && c != '_' && c != '-')
.unwrap_or(remaining.len());
(&remaining[..end], &remaining[end..])
};
labels.insert(key, value.to_string());
remaining = rest.trim_start_matches(',').trim_start();
} else {
break;
}
}
labels
}

217
apps/metrics/src/stats_store.rs
View File

@ -1,217 +0,0 @@
//! Stats store: receives expanded push payloads from all apps,
//! aggregates over time, computes derived statistics (p99 etc),
//! and provides JSON API for external consumption.
use serde::Serialize;
use std::collections::HashMap;
use std::sync::Arc;
use tokio::sync::RwLock;
/// Per-app, per-instance aggregated stats entry.
#[derive(Debug, Clone, Default, Serialize)]
pub struct AppStats {
/// Last seen timestamp.
pub last_seen: i64,
/// Number of push samples received.
pub sample_count: u64,
// ── HTTP ─────────────────────────────────────────────────────
pub requests_total: u64,
pub request_duration_ms_total: u64,
pub requests_2xx: u64,
pub requests_4xx: u64,
pub requests_5xx: u64,
pub endpoints: HashMap<String, u64>,
// ── System ───────────────────────────────────────────────────
pub cpu_usage_percent: f32,
pub memory_used_mb: u64,
pub memory_total_mb: u64,
pub uptime_secs: u64,
// ── Business counters ────────────────────────────────────────
pub business: HashMap<String, f64>,
// ── Token usage ──────────────────────────────────────────────
pub ai_input_tokens_total: i64,
pub ai_output_tokens_total: i64,
pub ai_calls_total: i64,
pub ai_calls_success: i64,
pub ai_calls_failure: i64,
pub token_by_model: HashMap<String, ModelTokenStats>,
// ── Tasks ────────────────────────────────────────────────────
pub tasks_queued: i64,
pub tasks_running: i64,
pub tasks_completed: i64,
pub tasks_failed: i64,
// ── Latency ──────────────────────────────────────────────────
pub latency: HashMap<String, LatencyStats>,
// ── Logs ─────────────────────────────────────────────────────
#[serde(skip_serializing)]
pub logs: Vec<(i64, String)>,
}
#[derive(Debug, Clone, Default, Serialize)]
pub struct ModelTokenStats {
pub input_tokens: i64,
pub output_tokens: i64,
pub calls: i64,
}
#[derive(Debug, Clone, Default, Serialize)]
pub struct LatencyStats {
pub p50_ms: f64,
pub p90_ms: f64,
pub p99_ms: f64,
pub max_ms: f64,
pub count: u64,
}
/// The global stats store: app_name → AppStats.
pub type StatsStore = Arc<RwLock<HashMap<String, AppStats>>>;
/// Merge a new push payload into the stats store.
pub async fn merge_push_payload(
store: &StatsStore,
app: &str,
_instance: &str,
timestamp: i64,
http: Option<&observability::push::HttpPayload>,
system: Option<&observability::push::SystemPayload>,
business: &HashMap<String, f64>,
token_usage: Option<&observability::push::TokenUsagePayload>,
tasks: Option<&observability::push::TaskStatsPayload>,
latency: &HashMap<String, observability::push::LatencySnapshot>,
logs: &[observability::push::LogEntry],
) {
// Use app_name as key (merge across instances for aggregation)
let mut guard = store.write().await;
let entry = guard.entry(app.to_string()).or_default();
entry.last_seen = timestamp;
entry.sample_count += 1;
// HTTP — accumulate (not replace, so we get totals over time)
if let Some(http) = http {
entry.requests_total = http.requests_total;
entry.request_duration_ms_total = http.request_duration_ms_total;
entry.requests_2xx = http.requests_2xx;
entry.requests_4xx = http.requests_4xx;
entry.requests_5xx = http.requests_5xx;
for (ep, count) in &http.endpoints {
*entry.endpoints.entry(ep.clone()).or_insert(0) = *count;
}
}
// System — replace (current snapshot, not cumulative)
if let Some(sys) = system {
entry.cpu_usage_percent = sys.cpu_usage_percent;
entry.memory_used_mb = sys.memory_used_mb;
entry.memory_total_mb = sys.memory_total_mb;
entry.uptime_secs = sys.uptime_secs;
}
// Business — replace with latest snapshot
entry.business = business.clone();
// Token usage — replace with latest
if let Some(tu) = token_usage {
entry.ai_input_tokens_total = tu.ai_input_tokens_total;
entry.ai_output_tokens_total = tu.ai_output_tokens_total;
entry.ai_calls_total = tu.ai_calls_total;
entry.ai_calls_success = tu.ai_calls_success;
entry.ai_calls_failure = tu.ai_calls_failure;
for (model, usage) in &tu.by_model {
let ms = entry.token_by_model.entry(model.clone()).or_default();
ms.input_tokens = usage.input_tokens;
ms.output_tokens = usage.output_tokens;
ms.calls = usage.calls;
}
}
// Tasks — replace with latest
if let Some(t) = tasks {
entry.tasks_queued = t.queued;
entry.tasks_running = t.running;
entry.tasks_completed = t.completed;
entry.tasks_failed = t.failed;
}
// Latency — replace with latest snapshots
for (endpoint, snap) in latency {
let ls = entry.latency.entry(endpoint.clone()).or_default();
ls.p50_ms = snap.p50_ms;
ls.p90_ms = snap.p90_ms;
ls.p99_ms = snap.p99_ms;
ls.max_ms = snap.max_ms;
ls.count = snap.count;
}
// Logs — append (keep last 300 lines)
for log in logs {
entry.logs.push((
log.timestamp,
format!("[{}] {}", log.level.to_lowercase(), log.message),
));
}
let cutoff = chrono::Utc::now().timestamp() - 300;
entry.logs.retain(|(ts, _)| *ts >= cutoff);
}
/// Dashboard response combining all apps' stats.
#[derive(Debug, Serialize)]
pub struct DashboardResponse {
/// Timestamp of this snapshot.
pub timestamp: i64,
/// Total number of app instances reporting.
pub app_count: u64,
/// Per-app aggregated stats.
pub apps: HashMap<String, AppStats>,
/// Derived: average p99 latency across all apps.
pub avg_p99_ms: f64,
/// Derived: total tokens consumed across all apps.
pub total_input_tokens: i64,
pub total_output_tokens: i64,
/// Derived: total AI calls across all apps.
pub total_ai_calls: i64,
}
/// Build the dashboard response from the stats store.
pub async fn build_dashboard(store: &StatsStore) -> DashboardResponse {
let guard = store.read().await;
let mut avg_p99 = 0.0;
let mut p99_count = 0;
let mut total_input = 0i64;
let mut total_output = 0i64;
let mut total_calls = 0i64;
for (_, stats) in guard.iter() {
total_input += stats.ai_input_tokens_total;
total_output += stats.ai_output_tokens_total;
total_calls += stats.ai_calls_total;
for (_, lat) in &stats.latency {
avg_p99 += lat.p99_ms;
p99_count += 1;
}
}
let avg_p99_ms = if p99_count > 0 {
avg_p99 / p99_count as f64
} else {
0.0
};
DashboardResponse {
timestamp: chrono::Utc::now().timestamp(),
app_count: guard.len() as u64,
apps: guard.clone(),
avg_p99_ms,
total_input_tokens: total_input,
total_output_tokens: total_output,
total_ai_calls: total_calls,
}
}

36
apps/metrics/src/target.rs
View File

@ -1,36 +0,0 @@
use anyhow::Context;
use serde::{Deserialize, Serialize};
use std::collections::HashMap;
#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct ScrapeTarget {
pub name: String,
pub addr: String,
#[serde(default = "default_metrics_path")]
pub metrics_path: String,
#[serde(default)]
pub labels: HashMap<String, String>,
}
fn default_metrics_path() -> String {
"/metrics".to_string()
}
impl ScrapeTarget {
pub fn url(&self) -> String {
if self.metrics_path.starts_with("http") {
self.metrics_path.clone()
} else {
format!("http://{}{}", self.addr, self.metrics_path)
}
}
}
pub async fn load_targets_from_file(path: &str) -> anyhow::Result<Vec<ScrapeTarget>> {
let content = tokio::fs::read_to_string(path)
.await
.context("read targets file")?;
let targets: Vec<ScrapeTarget> =
serde_json::from_str(&content).with_context(|| format!("parse targets file {path}"))?;
Ok(targets)
}

13
apps/migrate/Cargo.toml
View File

@ -1,13 +0,0 @@
[package]
name = "migrate-cli"
version.workspace = true
edition.workspace = true
[dependencies]
migrate.workspace = true
sea-orm = { workspace = true, features = ["sqlx-all", "runtime-tokio"] }
tokio = { workspace = true, features = ["rt-multi-thread", "macros"] }
anyhow.workspace = true
clap.workspace = true
dotenvy.workspace = true
config = { workspace = true }

102
apps/migrate/src/main.rs
View File

@ -1,102 +0,0 @@
use anyhow::Context;
use clap::Command;
use migrate::MigratorTrait;
use sea_orm::{Database, DatabaseConnection};
#[tokio::main]
async fn main() -> anyhow::Result<()> {
dotenvy::dotenv().ok();
config::AppConfig::load();
let cmd = Command::new("migrate")
.about("Database migration CLI")
.arg(
clap::Arg::new("steps")
.help("Number of migrations (for up/down)")
.required(false)
.index(1),
)
.subcommand(Command::new("up").about("Apply pending migrations"))
.subcommand(Command::new("down").about("Revert applied migrations"))
.subcommand(Command::new("fresh").about("Drop all tables and re-apply"))
.subcommand(Command::new("refresh").about("Revert all then re-apply"))
.subcommand(Command::new("reset").about("Revert all applied migrations"))
.subcommand(Command::new("status").about("Show migration status"))
.try_get_matches()
.map_err(|e| anyhow::anyhow!("{}", e))?;
let db_url = config::AppConfig::load().database_url()?;
let db: DatabaseConnection = Database::connect(&db_url).await?;
match cmd.subcommand_name() {
Some("up") => {
let steps = cmd
.get_one::<String>("steps")
.and_then(|s| s.parse().ok())
.unwrap_or(0);
run_up(&db, steps).await?;
}
Some("down") => {
let steps = cmd
.get_one::<String>("steps")
.and_then(|s| s.parse().ok())
.unwrap_or(1);
run_down(&db, steps).await?;
}
Some("fresh") => run_fresh(&db).await?,
Some("refresh") => run_refresh(&db).await?,
Some("reset") => run_reset(&db).await?,
Some("status") => run_status(&db).await?,
_ => {
eprintln!(
"Usage: migrate <command>\nCommands: up, down, fresh, refresh, reset, status"
);
std::process::exit(1);
}
}
Ok(())
}
async fn run_up(db: &DatabaseConnection, steps: u32) -> anyhow::Result<()> {
migrate::Migrator::up(db, if steps == 0 { None } else { Some(steps) })
.await
.context("failed to run migrations up")?;
Ok(())
}
async fn run_down(db: &DatabaseConnection, steps: u32) -> anyhow::Result<()> {
migrate::Migrator::down(db, Some(steps))
.await
.context("failed to run migrations down")?;
Ok(())
}
async fn run_fresh(db: &DatabaseConnection) -> anyhow::Result<()> {
migrate::Migrator::fresh(db)
.await
.context("failed to run migrations fresh")?;
Ok(())
}
async fn run_refresh(db: &DatabaseConnection) -> anyhow::Result<()> {
migrate::Migrator::refresh(db)
.await
.context("failed to run migrations refresh")?;
Ok(())
}
async fn run_reset(db: &DatabaseConnection) -> anyhow::Result<()> {
migrate::Migrator::reset(db)
.await
.context("failed to run migrations reset")?;
Ok(())
}
async fn run_status(db: &DatabaseConnection) -> anyhow::Result<()> {
migrate::Migrator::status(db)
.await
.context("failed to get migration status")?;
Ok(())
}

21
apps/static/Cargo.toml
View File

@ -1,21 +0,0 @@
[package]
name = "static-server"
version.workspace = true
edition.workspace = true
[dependencies]
actix-web = { workspace = true }
actix-files = { workspace = true }
actix-cors = { workspace = true }
observability = { workspace = true }
metrics-exporter-prometheus = "0.13"
tokio = { workspace = true, features = ["full"] }
futures = { workspace = true }
serde = { workspace = true }
serde_json = { workspace = true }
mime = { workspace = true }
mime_guess2 = { workspace = true }
slog = { workspace = true }
anyhow = { workspace = true }
env_logger = { workspace = true }
log = "0.4"

212
apps/static/src/main.rs
View File

@ -1,212 +0,0 @@
use actix_cors::Cors;
use actix_files::Files;
use actix_web::dev::{Service, ServiceRequest, ServiceResponse};
use actix_web::{App, HttpResponse, HttpServer, http::header, web};
use futures::future::LocalBoxFuture;
use log::info;
use observability::{HttpMetrics, init_tracing_subscriber, install_recorder, push::MetricsPusher};
use std::path::PathBuf;
use std::sync::Arc;
use std::task::{Context, Poll};
use std::time::Instant;
/// Static file server for avatar, blob, and other static files
/// Serves files from /data/{type} directories
#[derive(Clone)]
struct StaticConfig {
root: PathBuf,
cors_enabled: bool,
}
impl StaticConfig {
fn from_env() -> Self {
let root = std::env::var("STATIC_ROOT").unwrap_or_else(|_| "/data".to_string());
let cors = std::env::var("STATIC_CORS").unwrap_or_else(|_| "true".to_string());
Self {
root: PathBuf::from(root),
cors_enabled: cors == "true" || cors == "1",
}
}
fn ensure_dir(&self, name: &str) -> PathBuf {
let dir = self.root.join(name);
if !dir.exists() {
std::fs::create_dir_all(&dir).ok();
}
dir
}
}
async fn health() -> HttpResponse {
HttpResponse::Ok().json(serde_json::json!({
"status": "ok",
"service": "static-server"
}))
}
/// Custom middleware that logs requests except for noisy paths (health, metrics, static files).
struct RequestLogger;
impl<S, B> actix_web::dev::Transform<S, ServiceRequest> for RequestLogger
where
S: Service<ServiceRequest, Response = ServiceResponse<B>, Error = actix_web::Error>,
S::Future: 'static,
B: 'static,
{
type Response = ServiceResponse<B>;
type Error = actix_web::Error;
type Transform = RequestLoggerService<S>;
type InitError = ();
type Future = futures::future::Ready<Result<Self::Transform, Self::InitError>>;
fn new_transform(&self, service: S) -> Self::Future {
futures::future::ok(RequestLoggerService {
service,
_marker: std::marker::PhantomData,
})
}
}
struct RequestLoggerService<S> {
service: S,
_marker: std::marker::PhantomData<fn(ServiceRequest)>,
}
impl<S, B> Service<ServiceRequest> for RequestLoggerService<S>
where
S: Service<ServiceRequest, Response = ServiceResponse<B>, Error = actix_web::Error>,
S::Future: 'static,
B: 'static,
{
type Response = ServiceResponse<B>;
type Error = actix_web::Error;
type Future = LocalBoxFuture<'static, Result<Self::Response, Self::Error>>;
fn poll_ready(&self, cx: &mut Context<'_>) -> Poll<Result<(), Self::Error>> {
self.service.poll_ready(cx)
}
fn call(&self, req: ServiceRequest) -> Self::Future {
let path = req.path().to_string();
let method = req.method().to_string();
let should_log = !(path == "/health"
|| path == "/metrics"
|| path.starts_with("/ws")
|| path.starts_with("/avatar")
|| path.starts_with("/blob")
|| path.starts_with("/media")
|| path.starts_with("/static"));
let start = Instant::now();
let fut = self.service.call(req);
Box::pin(async move {
let res = fut.await?;
if should_log {
info!(
target: "static_server",
"{} {} {} {:?}",
method,
path,
res.status().as_u16(),
start.elapsed()
);
}
Ok(res)
})
}
}
#[actix_web::main]
async fn main() -> anyhow::Result<()> {
init_tracing_subscriber("info", false);
let prometheus_handle = Arc::new(install_recorder());
let http_metrics = Arc::new(HttpMetrics::new());
// Metrics pusher: periodically push all metrics to apps/metrics aggregator
if let Some(push_url) = std::env::var("METRICS_PUSH_URL").ok() {
let pusher = MetricsPusher::new(&push_url, "static");
pusher.spawn(
http_metrics.clone(),
prometheus_handle.clone(),
std::time::Duration::from_secs(15),
);
info!("Metrics pusher started (interval 15s, url: {})", push_url);
}
let cfg = StaticConfig::from_env();
let bind = std::env::var("STATIC_BIND").unwrap_or_else(|_| "0.0.0.0:8081".to_string());
println!("Static file server starting...");
println!(" Root: {:?}", cfg.root);
println!(" Bind: {}", bind);
println!(
" CORS: {}",
if cfg.cors_enabled {
"enabled"
} else {
"disabled"
}
);
// Ensure all directories exist
for name in ["avatar", "blob", "media", "static"] {
let dir = cfg.ensure_dir(name);
println!(" {} dir: {:?}", name, dir);
}
let root = cfg.root.clone();
let cors_enabled = cfg.cors_enabled;
HttpServer::new(move || {
let root = root.clone();
let cors = if cors_enabled {
// WARNING: allow_any_origin is intentional for static asset serving (CDN mode)
// Ensure no sensitive files are served from this directory
Cors::default()
.allow_any_origin()
.allowed_methods(vec!["GET", "HEAD", "OPTIONS"])
.allowed_headers(vec![
header::AUTHORIZATION,
header::ACCEPT,
header::CONTENT_TYPE,
])
.max_age(3600)
} else {
Cors::permissive()
};
App::new()
.wrap(cors)
.wrap(RequestLogger)
.route("/health", web::get().to(health))
.service(
Files::new("/avatar", root.join("avatar"))
.prefer_utf8(true)
.index_file("index.html"),
)
.service(
Files::new("/blob", root.join("blob"))
.prefer_utf8(true)
.index_file("index.html"),
)
.service(
Files::new("/media", root.join("media"))
.prefer_utf8(true)
.index_file("index.html"),
)
.service(
Files::new("/static", root.join("static"))
.prefer_utf8(true)
.index_file("index.html"),
)
})
.bind(&bind)?
.run()
.await?;
Ok(())
}

1807
bun.lock
View File

File diff suppressed because it is too large Load Diff

8
components.json
View File

@ -1,6 +1,6 @@
{ {
"$schema": "https://ui.shadcn.com/schema.json", "$schema": "https://ui.shadcn.com/schema.json",
"style": "radix-nova", "style": "base-nova",
"rsc": false, "rsc": false,
"tsx": true, "tsx": true,
"tailwind": { "tailwind": {
@ -12,6 +12,8 @@
}, },
"iconLibrary": "lucide", "iconLibrary": "lucide",
"rtl": false, "rtl": false,
"menuColor": "default",
"menuAccent": "subtle",
"aliases": { "aliases": {
"components": "@/components", "components": "@/components",
"utils": "@/lib/utils", "utils": "@/lib/utils",
@ -19,9 +21,7 @@
"lib": "@/lib", "lib": "@/lib",
"hooks": "@/hooks" "hooks": "@/hooks"
}, },
"menuColor": "default",
"menuAccent": "subtle",
"registries": { "registries": {
"@ai-elements": "https://ai-sdk.dev/elements/api/registry/{name}.json" "@manifest": "https://ui.manifest.build/r/{name}.json"
} }
} }

25
deploy/.helmignore
View File

@ -1,25 +0,0 @@
# Patterns to ignore when building packages.
# This supports shell glob matching, relative path matching, and
# negation (prefixed with !). Only one pattern per line.
.DS_Store
# Common VCS dirs
.git/
.gitignore
.bzr/
.bzrignore
.hg/
.hgignore
.svn/
# Common backup files
*.swp
*.bak
*.tmp
*.orig
*~
# Various IDEs
.project
.idea/
*.tmproj
.vscode/
# Secrets
.server.yaml

6
deploy/Chart.yaml
View File

@ -1,6 +0,0 @@
apiVersion: v2
name: deploy
description: Helm chart for the project backend services
type: application
version: 0.1.0
appVersion: "0.2.9"

209
deploy/README.md
View File

@ -1,209 +0,0 @@
# Deploy Helm Chart
Monolithic Helm chart for all backend services.
## Services
| Service | Port(s) | Replicas | HPA | Purpose |
|----------------------|-------------------------|----------|----------|---------------------------------------------|
| `app` | 3000 (HTTP) | 2 | 2–10 | Main API server |
| `gitserver` | 8021 (HTTP), 2222 (SSH) | 1 | 1–5 | Git HTTP + SSH server |
| `email_worker` | 8084 (HTTP) | 1 | disabled | Email queue consumer (single instance only) |
| `git_hook` | 8083 (HTTP) | 1 | 1–5 | Git hook worker pool |
| `metrics_aggregator` | 9090 (HTTP) | 1 | 1–5 | Prometheus scrape + Loki push |
| `static_server` | 8081 (HTTP) | 1 | 1–5 | Static file server (avatars, blobs, media) |
## Prerequisites
The following resources must exist in the cluster **before** installing the Helm chart. They are not managed by Helm —
install, upgrade, and uninstall of the chart will not touch them.
### 1. Namespace
```bash
kubectl create namespace app
```
### 2. PVC (aliyun-nfs-app, 200Ti, ReadWriteMany)
```bash
kubectl apply -f - <<'EOF'
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
name: shared-data
namespace: app
spec:
accessModes:
- ReadWriteMany
resources:
requests:
storage: 200Ti
storageClassName: aliyun-nfs-app
EOF
```
> The chart references this PVC by hardcoded name `shared-data`. This name is immutable — it cannot be changed via Helm
> values.
### 3. ConfigMap
```bash
kubectl apply -f - <<'EOF'
apiVersion: v1
kind: ConfigMap
metadata:
name: app-env
namespace: app
data:
APP_REPOS_ROOT: "/data/repos"
APP_AVATAR_PATH: "/data/avatars"
STORAGE_PATH: "/data/files"
STATIC_ROOT: "/data"
APP_LOG_LEVEL: "info"
APP_COOKIE_SECURE: "false"
APP_DOMAIN_URL: "https://your-domain.com"
APP_DATABASE_URL: "postgres://user:pass@postgres:5432/app"
APP_REDIS_URL: "redis://redis:6379"
APP_AI_BASIC_URL: "https://api.openai.com/v1"
APP_AI_API_KEY: "sk-..."
APP_SMTP_PASSWORD: "..."
APP_SESSION_SECRET: "min-32-byte-random-string..."
APP_SSH_SERVER_PRIVATE_KEY: "<hex-encoded-private-key>"
EOF
```
| Variable | Default / Example | Required |
|------------------------------|-----------------------------|-----------|
| `APP_REPOS_ROOT` | `/data/repos` | Yes |
| `APP_AVATAR_PATH` | `/data/avatars` | Yes |
| `STORAGE_PATH` | `/data/files` | Yes |
| `STATIC_ROOT` | `/data` | Yes |
| `APP_LOG_LEVEL` | `info` | No |
| `APP_COOKIE_SECURE` | `false` | No |
| `APP_DOMAIN_URL` | `https://your-domain.com` | Yes |
| `APP_DATABASE_URL` | `postgres://...` | **Yes** |
| `APP_REDIS_URL` | `redis://...` | **Yes** |
| `APP_AI_BASIC_URL` | `https://api.openai.com/v1` | **Yes** |
| `APP_AI_API_KEY` | `sk-...` | **Yes** |
| `APP_SMTP_PASSWORD` | `...` | **Yes** |
| `APP_SESSION_SECRET` | min 32 bytes | **Yes** |
| `APP_SSH_SERVER_PRIVATE_KEY` | hex-encoded PEM | **Yes** |
| `APP_SSH_PORT` | `2222` | Yes (k8s) |
> **SSH host key**: `APP_SSH_SERVER_PRIVATE_KEY` must be the hex-encoded Ed25519 private key PEM bytes.
> ```bash
> ssh-keygen -t ed25519 -f /tmp/ssh_host_key -N ""
> hexdump -v -e '/1 "%02x"' < /tmp/ssh_host_key
> ```
>
> **Session secret**: generate 48 random bytes:
> ```bash
> openssl rand -base64 48
> ```
>
> Override the ConfigMap name with `--set configMapName=your-cm-name`.
### 4. Verify prerequisites
```bash
kubectl get namespace app
kubectl get pvc -n app shared-data
kubectl get configmap -n app app-env
```
## Quick Start
```bash
helm template deploy ./deploy --namespace app --set imageRegistry=ghcr.io/your-org
helm lint ./deploy
# Install
helm upgrade --install deploy ./deploy \
--namespace app \
--set imageRegistry=ghcr.io/your-org \
--set imageTag=v0.2.9
```
## Storage
All services share a single PVC (`shared-data`) via `subPath` mounts:
| SubPath | Mount | Used By |
|-----------|-----------------|--------------------------|
| `repos` | `/data/repos` | app, gitserver, git-hook |
| `avatars` | `/data/avatars` | app |
| `files` | `/data/files` | app |
| `static` | `/data` | static-server |
Pods run as UID/GID `1000` and set `fsGroup: 1000` so Git processes can create temporary object
directories under bare repositories. If an existing PVC was previously written by another UID,
fix ownership once from a maintenance pod:
```bash
chown -R 1000:1000 /data/repos
chmod -R u+rwX,g+rwX /data/repos
```
## Autoscaling
All services except `email_worker` have HPA enabled by default. The email worker is fixed at 1 replica and must not be
scaled.
To adjust HPA bounds per service:
```bash
--set services.app.autoscaling.maxReplicas=20
--set services.app.autoscaling.targetCPUUtilization=70
```
To disable HPA for a service:
```bash
--set services.git_hook.autoscaling.enabled=false
```
## Ingress
```bash
helm upgrade --install deploy ./deploy \
--namespace app \
--set ingress.enabled=true \
--set ingress.className=nginx \
--set ingress.hosts[0].host=your-domain.com
```
## Dependencies
All services require these to be reachable from the cluster:
- PostgreSQL (via `APP_DATABASE_URL`)
- Redis (via `APP_REDIS_URL`)
- Git binary (included in all Docker images)
- OpenAI-compatible API (via `APP_AI_BASIC_URL` + `APP_AI_API_KEY`)
- Qdrant vector DB (via `APP_QDRANT_URL`)
- SMTP server (via `APP_SMTP_*`)
- Embedding model (via `APP_EMBED_MODEL_*`)
Optional dependencies with graceful degradation:
| Dependency | Variable | Fallback |
|----------------|-------------------------------|------------------|
| NATS JetStream | `NATS_URL` + `NATS_TOKEN` | Redis queue |
| Loki | `LOKI_URL` | Logs discarded |
| OTEL Collector | `OTEL_EXPORTER_OTLP_ENDPOINT` | Tracing disabled |
## Production Example
```bash
helm upgrade --install deploy ./deploy \
--namespace app \
--set imageRegistry=ghcr.io/your-org \
--set imageTag=v0.2.9 \
--set services.app.replicas=3 \
--set services.app.autoscaling.maxReplicas=20 \
--set ingress.enabled=true \
--set ingress.className=nginx \
--set ingress.hosts[0].host=your-domain.com \
--set configMapName=app-env
```

19
deploy/templates/NOTES.txt
View File

@ -1,19 +0,0 @@
Project backend services deployed to namespace: {{ .Release.Namespace }}
Services:
{{- range $svcKey, $svcVal := .Values.services }}
{{ $svcKey | replace "_" "-" }}: {{ if $svcVal.ports }}{{ range $portName, $portNum := $svcVal.ports }}{{ $portName }}={{ $portNum }} {{ end }}{{ else }}port={{ $svcVal.port }}{{ end }} {{ if $svcVal.autoscaling.enabled }}(HPA: {{ $svcVal.autoscaling.minReplicas }}-{{ $svcVal.autoscaling.maxReplicas }}){{ else }}(static: {{ $svcVal.replicaCount }}){{ end }}
{{- end }}
To access the app locally:
kubectl port-forward -n {{ .Release.Namespace }} svc/{{ include "deploy.serviceFullname" (dict "root" . "svcKey" "app") }} 3000:3000
To check HPA status:
{{- range $svcKey, $svcVal := .Values.services }}
{{- if $svcVal.autoscaling.enabled }}
kubectl get hpa -n {{ $.Release.Namespace }} {{ include "deploy.serviceFullname" (dict "root" $ "svcKey" $svcKey) }}
{{- end }}
{{- end }}
To check all pods:
kubectl get pods -n {{ .Release.Namespace }} -l "app.kubernetes.io/name={{ include "deploy.name" . }}"

78
deploy/templates/_helpers.tpl
View File

@ -1,78 +0,0 @@
{{/*
Expand the name of the chart.
*/}}
{{- define "deploy.name" -}}
{{- default .Chart.Name .Values.nameOverride | trunc 63 | trimSuffix "-" }}
{{- end }}
{{/*
Create a default fully qualified app name.
*/}}
{{- define "deploy.fullname" -}}
{{- if .Values.fullnameOverride }}
{{- .Values.fullnameOverride | trunc 63 | trimSuffix "-" }}
{{- else }}
{{- $name := default .Chart.Name .Values.nameOverride }}
{{- if contains $name .Release.Name }}
{{- .Release.Name | trunc 63 | trimSuffix "-" }}
{{- else }}
{{- printf "%s-%s" .Release.Name $name | trunc 63 | trimSuffix "-" }}
{{- end }}
{{- end }}
{{- end }}
{{/*
Service fullname — includes service key for per-service resources.
Underscores in svcKey are replaced with hyphens for valid Kubernetes names.
*/}}
{{- define "deploy.serviceFullname" -}}
{{- printf "%s-%s" (include "deploy.fullname" .root) (.svcKey | replace "_" "-") | trunc 63 | trimSuffix "-" }}
{{- end }}
{{/*
Chart name and version as used by the chart label.
*/}}
{{- define "deploy.chart" -}}
{{- printf "%s-%s" .Chart.Name .Chart.Version | replace "+" "_" | trunc 63 | trimSuffix "-" }}
{{- end }}
{{/*
Common labels
*/}}
{{- define "deploy.labels" -}}
helm.sh/chart: {{ include "deploy.chart" . }}
{{ include "deploy.selectorLabels" . }}
{{- if .Chart.AppVersion }}
app.kubernetes.io/version: {{ .Chart.AppVersion | quote }}
{{- end }}
app.kubernetes.io/managed-by: {{ .Release.Service }}
{{- end }}
{{/*
Selector labels
*/}}
{{- define "deploy.selectorLabels" -}}
app.kubernetes.io/name: {{ include "deploy.name" . }}
app.kubernetes.io/instance: {{ .Release.Name }}
{{- end }}
{{/*
Per-service selector labels — used by Service to target the right Deployment.
Underscores in svcKey are replaced with hyphens for valid Kubernetes label values.
*/}}
{{- define "deploy.serviceSelectorLabels" -}}
app.kubernetes.io/name: {{ include "deploy.name" .root }}
app.kubernetes.io/instance: {{ .root.Release.Name }}
app.kubernetes.io/component: {{ .svcKey | replace "_" "-" }}
{{- end }}
{{/*
Create the name of the service account to use
*/}}
{{- define "deploy.serviceAccountName" -}}
{{- if .Values.serviceAccount.create }}
{{- default (include "deploy.fullname" .) .Values.serviceAccount.name }}
{{- else }}
{{- default "default" .Values.serviceAccount.name }}
{{- end }}
{{- end }}

89
deploy/templates/app/deployment.yaml
View File

@ -1,89 +0,0 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "app") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: app
spec:
replicas: {{ .Values.services.app.replicaCount | default 1 }}
selector:
matchLabels:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "app") | nindent 6 }}
template:
metadata:
labels:
{{- include "deploy.labels" . | nindent 8 }}
app.kubernetes.io/component: app
spec:
{{- with .Values.imagePullSecrets }}
imagePullSecrets:
{{- toYaml . | nindent 8 }}
{{- end }}
serviceAccountName: {{ include "deploy.serviceAccountName" . }}
{{- with .Values.podSecurityContext }}
securityContext:
{{- toYaml . | nindent 8 }}
{{- end }}
containers:
- name: app
{{- with .Values.securityContext }}
securityContext:
{{- toYaml . | nindent 12 }}
{{- end }}
image: "{{ .Values.imageRegistry }}/{{ .Values.services.app.repository }}:{{ .Values.imageTag | default .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
{{- with .Values.services.app.command }}
command:
{{- toYaml . | nindent 12 }}
{{- end }}
ports:
- name: http
containerPort: {{ .Values.services.app.port }}
protocol: TCP
envFrom:
- configMapRef:
name: {{ .Values.configMapName }}
{{- with .Values.services.app.extraEnv }}
env:
{{- range $key, $val := . }}
- name: {{ $key }}
value: {{ $val | quote }}
{{- end }}
{{- end }}
livenessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 10
periodSeconds: 15
readinessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 5
periodSeconds: 10
{{- with .Values.services.app.resources }}
resources:
{{- toYaml . | nindent 12 }}
{{- end }}
{{- with .Values.services.app.volumeMounts }}
volumeMounts:
{{- toYaml . | nindent 12 }}
{{- end }}
volumes:
- name: shared-data
persistentVolumeClaim:
claimName: shared-data
{{- with .Values.nodeSelector }}
nodeSelector:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.affinity }}
affinity:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.tolerations }}
tolerations:
{{- toYaml . | nindent 8 }}
{{- end }}

16
deploy/templates/app/service.yaml
View File

@ -1,16 +0,0 @@
apiVersion: v1
kind: Service
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "app") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: app
spec:
type: ClusterIP
ports:
- port: {{ .Values.services.app.port }}
targetPort: http
protocol: TCP
name: http
selector:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "app") | nindent 4 }}

70
deploy/templates/email_worker/deployment.yaml
View File

@ -1,70 +0,0 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "email_worker") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: email-worker
spec:
replicas: {{ .Values.services.email_worker.replicaCount | default 1 }}
selector:
matchLabels:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "email_worker") | nindent 6 }}
template:
metadata:
labels:
{{- include "deploy.labels" . | nindent 8 }}
app.kubernetes.io/component: email-worker
spec:
{{- with .Values.imagePullSecrets }}
imagePullSecrets:
{{- toYaml . | nindent 8 }}
{{- end }}
serviceAccountName: {{ include "deploy.serviceAccountName" . }}
{{- with .Values.podSecurityContext }}
securityContext:
{{- toYaml . | nindent 8 }}
{{- end }}
containers:
- name: email-worker
{{- with .Values.securityContext }}
securityContext:
{{- toYaml . | nindent 12 }}
{{- end }}
image: "{{ .Values.imageRegistry }}/{{ .Values.services.email_worker.repository }}:{{ .Values.imageTag | default .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: {{ .Values.services.email_worker.port }}
protocol: TCP
envFrom:
- configMapRef:
name: {{ .Values.configMapName }}
livenessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 10
periodSeconds: 15
readinessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 5
periodSeconds: 10
{{- with .Values.services.email_worker.resources }}
resources:
{{- toYaml . | nindent 12 }}
{{- end }}
{{- with .Values.nodeSelector }}
nodeSelector:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.affinity }}
affinity:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.tolerations }}
tolerations:
{{- toYaml . | nindent 8 }}
{{- end }}

16
deploy/templates/email_worker/service.yaml
View File

@ -1,16 +0,0 @@
apiVersion: v1
kind: Service
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "email_worker") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: email-worker
spec:
type: ClusterIP
ports:
- port: {{ .Values.services.email_worker.port }}
targetPort: http
protocol: TCP
name: http
selector:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "email_worker") | nindent 4 }}

78
deploy/templates/git_hook/deployment.yaml
View File

@ -1,78 +0,0 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "git_hook") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: git-hook
spec:
replicas: {{ .Values.services.git_hook.replicaCount | default 1 }}
selector:
matchLabels:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "git_hook") | nindent 6 }}
template:
metadata:
labels:
{{- include "deploy.labels" . | nindent 8 }}
app.kubernetes.io/component: git-hook
spec:
{{- with .Values.imagePullSecrets }}
imagePullSecrets:
{{- toYaml . | nindent 8 }}
{{- end }}
serviceAccountName: {{ include "deploy.serviceAccountName" . }}
{{- with .Values.podSecurityContext }}
securityContext:
{{- toYaml . | nindent 8 }}
{{- end }}
containers:
- name: git-hook
{{- with .Values.securityContext }}
securityContext:
{{- toYaml . | nindent 12 }}
{{- end }}
image: "{{ .Values.imageRegistry }}/{{ .Values.services.git_hook.repository }}:{{ .Values.imageTag | default .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: {{ .Values.services.git_hook.port }}
protocol: TCP
envFrom:
- configMapRef:
name: {{ .Values.configMapName }}
livenessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 10
periodSeconds: 15
readinessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 5
periodSeconds: 10
{{- with .Values.services.git_hook.resources }}
resources:
{{- toYaml . | nindent 12 }}
{{- end }}
{{- with .Values.services.git_hook.volumeMounts }}
volumeMounts:
{{- toYaml . | nindent 12 }}
{{- end }}
volumes:
- name: shared-data
persistentVolumeClaim:
claimName: shared-data
{{- with .Values.nodeSelector }}
nodeSelector:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.affinity }}
affinity:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.tolerations }}
tolerations:
{{- toYaml . | nindent 8 }}
{{- end }}

16
deploy/templates/git_hook/service.yaml
View File

@ -1,16 +0,0 @@
apiVersion: v1
kind: Service
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "git_hook") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: git-hook
spec:
type: ClusterIP
ports:
- port: {{ .Values.services.git_hook.port }}
targetPort: http
protocol: TCP
name: http
selector:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "git_hook") | nindent 4 }}

88
deploy/templates/gitserver/deployment.yaml
View File

@ -1,88 +0,0 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "gitserver") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: gitserver
spec:
replicas: {{ .Values.services.gitserver.replicaCount | default 1 }}
selector:
matchLabels:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "gitserver") | nindent 6 }}
template:
metadata:
labels:
{{- include "deploy.labels" . | nindent 8 }}
app.kubernetes.io/component: gitserver
spec:
{{- with .Values.imagePullSecrets }}
imagePullSecrets:
{{- toYaml . | nindent 8 }}
{{- end }}
serviceAccountName: {{ include "deploy.serviceAccountName" . }}
{{- with .Values.podSecurityContext }}
securityContext:
{{- toYaml . | nindent 8 }}
{{- end }}
containers:
- name: gitserver
{{- with .Values.securityContext }}
securityContext:
{{- toYaml . | nindent 12 }}
{{- end }}
image: "{{ .Values.imageRegistry }}/{{ .Values.services.gitserver.repository }}:{{ .Values.imageTag | default .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: {{ .Values.services.gitserver.ports.http }}
protocol: TCP
- name: ssh
containerPort: {{ .Values.services.gitserver.ports.ssh }}
protocol: TCP
envFrom:
- configMapRef:
name: {{ .Values.configMapName }}
{{- with .Values.services.gitserver.extraEnv }}
env:
{{- range $key, $val := . }}
- name: {{ $key }}
value: {{ $val | quote }}
{{- end }}
{{- end }}
livenessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 10
periodSeconds: 15
readinessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 5
periodSeconds: 10
{{- with .Values.services.gitserver.resources }}
resources:
{{- toYaml . | nindent 12 }}
{{- end }}
{{- with .Values.services.gitserver.volumeMounts }}
volumeMounts:
{{- toYaml . | nindent 12 }}
{{- end }}
volumes:
- name: shared-data
persistentVolumeClaim:
claimName: shared-data
{{- with .Values.nodeSelector }}
nodeSelector:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.affinity }}
affinity:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.tolerations }}
tolerations:
{{- toYaml . | nindent 8 }}
{{- end }}

20
deploy/templates/gitserver/service.yaml
View File

@ -1,20 +0,0 @@
apiVersion: v1
kind: Service
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "gitserver") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: gitserver
spec:
type: ClusterIP
ports:
- port: {{ .Values.services.gitserver.ports.http }}
targetPort: http
protocol: TCP
name: http
- port: {{ .Values.services.gitserver.ports.ssh }}
targetPort: ssh
protocol: TCP
name: ssh
selector:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "gitserver") | nindent 4 }}

21
deploy/templates/gitserver/ssh-service.yaml
View File

@ -1,21 +0,0 @@
apiVersion: v1
kind: Service
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "gitserver") }}-ssh
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: gitserver
annotations:
{{- with .Values.services.gitserver.sshService.annotations }}
{{- toYaml . | nindent 4 }}
{{- end }}
spec:
type: LoadBalancer
externalTrafficPolicy: Local
ports:
- port: 22
targetPort: ssh
protocol: TCP
name: ssh
selector:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "gitserver") | nindent 4 }}

26
deploy/templates/hpa.yaml
View File

@ -1,26 +0,0 @@
{{- range $svcKey, $svcVal := .Values.services }}
{{- if $svcVal.autoscaling.enabled }}
---
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" $ "svcKey" $svcKey) }}
labels:
{{- include "deploy.labels" $ | nindent 4 }}
app.kubernetes.io/component: {{ $svcKey | replace "_" "-" }}
spec:
scaleTargetRef:
apiVersion: apps/v1
kind: Deployment
name: {{ include "deploy.serviceFullname" (dict "root" $ "svcKey" $svcKey) }}
minReplicas: {{ $svcVal.autoscaling.minReplicas }}
maxReplicas: {{ $svcVal.autoscaling.maxReplicas }}
metrics:
- type: Resource
resource:
name: cpu
target:
type: Utilization
averageUtilization: {{ $svcVal.autoscaling.targetCPUUtilization }}
{{- end }}
{{- end }}

41
deploy/templates/ingress.yaml
View File

@ -1,41 +0,0 @@
{{- if .Values.ingress.enabled -}}
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
name: {{ include "deploy.fullname" . }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
{{- with .Values.ingress.annotations }}
annotations:
{{- toYaml . | nindent 4 }}
{{- end }}
spec:
{{- with .Values.ingress.className }}
ingressClassName: {{ . }}
{{- end }}
{{- if .Values.ingress.tls }}
tls:
{{- range .Values.ingress.tls }}
- hosts:
{{- range .hosts }}
- {{ . | quote }}
{{- end }}
secretName: {{ .secretName }}
{{- end }}
{{- end }}
rules:
{{- range .Values.ingress.hosts }}
- host: {{ .host | quote }}
http:
paths:
{{- range .paths }}
- path: {{ .path }}
pathType: {{ .pathType }}
backend:
service:
name: {{ include "deploy.serviceFullname" (dict "root" $ "svcKey" .serviceName) }}
port:
number: {{ .servicePort }}
{{- end }}
{{- end }}
{{- end }}

70
deploy/templates/metrics_aggregator/deployment.yaml
View File

@ -1,70 +0,0 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "metrics_aggregator") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: metrics-aggregator
spec:
replicas: {{ .Values.services.metrics_aggregator.replicaCount | default 1 }}
selector:
matchLabels:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "metrics_aggregator") | nindent 6 }}
template:
metadata:
labels:
{{- include "deploy.labels" . | nindent 8 }}
app.kubernetes.io/component: metrics-aggregator
spec:
{{- with .Values.imagePullSecrets }}
imagePullSecrets:
{{- toYaml . | nindent 8 }}
{{- end }}
serviceAccountName: {{ include "deploy.serviceAccountName" . }}
{{- with .Values.podSecurityContext }}
securityContext:
{{- toYaml . | nindent 8 }}
{{- end }}
containers:
- name: metrics-aggregator
{{- with .Values.securityContext }}
securityContext:
{{- toYaml . | nindent 12 }}
{{- end }}
image: "{{ .Values.imageRegistry }}/{{ .Values.services.metrics_aggregator.repository }}:{{ .Values.imageTag | default .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: {{ .Values.services.metrics_aggregator.port }}
protocol: TCP
envFrom:
- configMapRef:
name: {{ .Values.configMapName }}
livenessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 10
periodSeconds: 15
readinessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 5
periodSeconds: 10
{{- with .Values.services.metrics_aggregator.resources }}
resources:
{{- toYaml . | nindent 12 }}
{{- end }}
{{- with .Values.nodeSelector }}
nodeSelector:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.affinity }}
affinity:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.tolerations }}
tolerations:
{{- toYaml . | nindent 8 }}
{{- end }}

16
deploy/templates/metrics_aggregator/service.yaml
View File

@ -1,16 +0,0 @@
apiVersion: v1
kind: Service
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "metrics_aggregator") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: metrics-aggregator
spec:
type: ClusterIP
ports:
- port: {{ .Values.services.metrics_aggregator.port }}
targetPort: http
protocol: TCP
name: http
selector:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "metrics_aggregator") | nindent 4 }}

1
deploy/templates/secret.yaml
View File

@ -1 +0,0 @@
{{/* Secret disabled — all config via ConfigMap */}}

13
deploy/templates/serviceaccount.yaml
View File

@ -1,13 +0,0 @@
{{- if .Values.serviceAccount.create -}}
apiVersion: v1
kind: ServiceAccount
metadata:
name: {{ include "deploy.serviceAccountName" . }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
{{- with .Values.serviceAccount.annotations }}
annotations:
{{- toYaml . | nindent 4 }}
{{- end }}
automountServiceAccountToken: {{ .Values.serviceAccount.automount }}
{{- end }}

78
deploy/templates/static_server/deployment.yaml
View File

@ -1,78 +0,0 @@
apiVersion: apps/v1
kind: Deployment
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "static_server") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: static-server
spec:
replicas: {{ .Values.services.static_server.replicaCount | default 1 }}
selector:
matchLabels:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "static_server") | nindent 6 }}
template:
metadata:
labels:
{{- include "deploy.labels" . | nindent 8 }}
app.kubernetes.io/component: static-server
spec:
{{- with .Values.imagePullSecrets }}
imagePullSecrets:
{{- toYaml . | nindent 8 }}
{{- end }}
serviceAccountName: {{ include "deploy.serviceAccountName" . }}
{{- with .Values.podSecurityContext }}
securityContext:
{{- toYaml . | nindent 8 }}
{{- end }}
containers:
- name: static-server
{{- with .Values.securityContext }}
securityContext:
{{- toYaml . | nindent 12 }}
{{- end }}
image: "{{ .Values.imageRegistry }}/{{ .Values.services.static_server.repository }}:{{ .Values.imageTag | default .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
ports:
- name: http
containerPort: {{ .Values.services.static_server.port }}
protocol: TCP
envFrom:
- configMapRef:
name: {{ .Values.configMapName }}
livenessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 10
periodSeconds: 15
readinessProbe:
httpGet:
path: /health
port: http
initialDelaySeconds: 5
periodSeconds: 10
{{- with .Values.services.static_server.resources }}
resources:
{{- toYaml . | nindent 12 }}
{{- end }}
{{- with .Values.services.static_server.volumeMounts }}
volumeMounts:
{{- toYaml . | nindent 12 }}
{{- end }}
volumes:
- name: shared-data
persistentVolumeClaim:
claimName: shared-data
{{- with .Values.nodeSelector }}
nodeSelector:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.affinity }}
affinity:
{{- toYaml . | nindent 8 }}
{{- end }}
{{- with .Values.tolerations }}
tolerations:
{{- toYaml . | nindent 8 }}
{{- end }}

16
deploy/templates/static_server/service.yaml
View File

@ -1,16 +0,0 @@
apiVersion: v1
kind: Service
metadata:
name: {{ include "deploy.serviceFullname" (dict "root" . "svcKey" "static_server") }}
labels:
{{- include "deploy.labels" . | nindent 4 }}
app.kubernetes.io/component: static-server
spec:
type: ClusterIP
ports:
- port: {{ .Values.services.static_server.port }}
targetPort: http
protocol: TCP
name: http
selector:
{{- include "deploy.serviceSelectorLabels" (dict "root" . "svcKey" "static_server") | nindent 4 }}

212
deploy/values.yaml
View File

@ -1,212 +0,0 @@
# Global image registry and tag
imageRegistry: ""
imageTag: ""
# External ConfigMap (managed outside Helm)
configMapName: "app-env"
# Service definitions
services:
app:
repository: app
port: 3000
replicaCount: 2
autoscaling:
enabled: true
minReplicas: 2
maxReplicas: 10
targetCPUUtilization: 80
command:
- "app"
- "--bind"
- "0.0.0.0:3000"
resources:
requests:
cpu: 200m
memory: 256Mi
limits:
cpu: "1"
memory: 512Mi
volumeMounts:
- name: shared-data
mountPath: /data/repos
subPath: repos
- name: shared-data
mountPath: /data/avatars
subPath: avatars
- name: shared-data
mountPath: /data/files
subPath: files
email_worker:
repository: email-worker
port: 8084
replicaCount: 1
autoscaling:
enabled: false # email must stay at 1 replica
resources:
requests:
cpu: 100m
memory: 128Mi
limits:
cpu: 500m
memory: 256Mi
git_hook:
repository: git-hook
port: 8083
replicaCount: 1
autoscaling:
enabled: true
minReplicas: 1
maxReplicas: 5
targetCPUUtilization: 80
resources:
requests:
cpu: 100m
memory: 128Mi
limits:
cpu: 500m
memory: 256Mi
volumeMounts:
- name: shared-data
mountPath: /data/repos
subPath: repos
gitserver:
repository: gitserver
ports:
http: 8021
ssh: 2222
replicaCount: 1
autoscaling:
enabled: true
minReplicas: 1
maxReplicas: 5
targetCPUUtilization: 80
# SSH port must match the containerPort
extraEnv:
APP_SSH_PORT: "2222"
# SSH service config (MetalLB + Cilium)
# Shared IP: nginx ingress (80/443) + SSH (22) on same VIP
# Requires ingress-nginx svc also annotated with allow-shared-ip: "gitdata-shared"
sshService:
annotations: {}
resources:
requests:
cpu: 100m
memory: 128Mi
limits:
cpu: 500m
memory: 256Mi
volumeMounts:
- name: shared-data
mountPath: /data/repos
subPath: repos
metrics_aggregator:
repository: metrics-aggregator
port: 9090
replicaCount: 1
autoscaling:
enabled: true
minReplicas: 1
maxReplicas: 5
targetCPUUtilization: 80
resources:
requests:
cpu: 100m
memory: 128Mi
limits:
cpu: 500m
memory: 256Mi
static_server:
repository: static-server
port: 8081
replicaCount: 1
autoscaling:
enabled: true
minReplicas: 1
maxReplicas: 5
targetCPUUtilization: 80
resources:
requests:
cpu: 50m
memory: 64Mi
limits:
cpu: 200m
memory: 128Mi
volumeMounts:
- name: shared-data
mountPath: /data
subPath: static
# Ingress
ingress:
enabled: true
className: "nginx"
annotations:
cert-manager.io/cluster-issuer: "cloudflare-acme-cluster-issuer"
nginx.ingress.kubernetes.io/proxy-body-size: "0"
nginx.ingress.kubernetes.io/proxy-read-timeout: "3600"
nginx.ingress.kubernetes.io/proxy-send-timeout: "3600"
nginx.ingress.kubernetes.io/affinity: "cookie"
nginx.ingress.kubernetes.io/session-cookie-name: "INGRESSROUTE"
nginx.ingress.kubernetes.io/session-cookie-path: "/"
nginx.ingress.kubernetes.io/session-cookie-max-age: "86400"
nginx.ingress.kubernetes.io/enable-real-ip: "true"
nginx.ingress.kubernetes.io/real-ip-header: "X-Forwarded-For"
nginx.ingress.kubernetes.io/use-forwarded-headers: "true"
hosts:
- host: gitdata.ai
paths:
- path: /
pathType: Prefix
serviceName: app
servicePort: 3000
- host: static.gitdata.ai
paths:
- path: /
pathType: Prefix
serviceName: static_server
servicePort: 8081
- host: git.gitdata.ai
paths:
- path: /
pathType: Prefix
serviceName: gitserver
servicePort: 8021
tls:
- secretName: gitdata-ai-tls
hosts:
- gitdata.ai
- static.gitdata.ai
- git.gitdata.ai
imagePullSecrets: []
nameOverride: ""
fullnameOverride: ""
serviceAccount:
create: true
automount: true
annotations: {}
name: ""
podSecurityContext:
runAsNonRoot: true
runAsUser: 1000
runAsGroup: 1000
fsGroup: 1000
fsGroupChangePolicy: OnRootMismatch
securityContext:
capabilities:
drop:
- ALL
readOnlyRootFilesystem: false
nodeSelector: {}
tolerations: []
affinity: {}

9
docker/app.Dockerfile
View File

@ -1,9 +0,0 @@
FROM ubuntu:24.04
RUN apt-get update && apt-get install -y --no-install-recommends \
ca-certificates libssl3 openssh-client procps git \
&& rm -rf /var/lib/apt/lists/*
RUN git config --system --add safe.directory '*'
WORKDIR /app
COPY ./target/release/app /bin
EXPOSE 3000
CMD ["app"]

65
docker/email.Dockerfile
View File

@ -1,8 +1,61 @@
FROM ubuntu:24.04 # GitDataAI Backend - Email Service
RUN apt-get update && apt-get install -y --no-install-recommends \ # Multi-stage build for Rust application
ca-certificates libssl3 \
# Stage 1: Build the application
FROM rust:1.96-bookworm AS builder
# Install system dependencies
RUN apt-get update && apt-get install -y \
pkg-config \
libssl-dev \
libpq-dev \
cmake \
&& rm -rf /var/lib/apt/lists/* && rm -rf /var/lib/apt/lists/*
# Create app directory
WORKDIR /app WORKDIR /app
COPY ./target/release/email-worker /bin
EXPOSE 8084 # Copy workspace files
CMD ["email-worker"] COPY Cargo.toml Cargo.lock ./
COPY app/ app/
COPY lib/ lib/
# Build the application in release mode
RUN cargo build --release --bin email-service
# Stage 2: Create runtime image
FROM debian:bookworm-slim
# Install runtime dependencies
RUN apt-get update && apt-get install -y \
libssl3 \
libpq5 \
ca-certificates \
curl \
&& rm -rf /var/lib/apt/lists/*
# Create non-root user
RUN useradd -r -s /bin/false appuser
# Create directories
RUN mkdir -p /app/logs \
&& chown -R appuser:appuser /app
# Copy binary from builder
COPY --from=builder /app/target/release/email-service /app/email-service
# Set ownership
RUN chown -R appuser:appuser /app
# Switch to non-root user
USER appuser
# Set working directory
WORKDIR /app
# Health check
HEALTHCHECK --interval=30s --timeout=3s --start-period=5s --retries=3 \
CMD pgrep email-service || exit 1
# Run the application
CMD ["./email-service"]

9
docker/githook.Dockerfile
View File

@ -1,9 +0,0 @@
FROM ubuntu:24.04
RUN apt-get update && apt-get install -y --no-install-recommends \
ca-certificates libssl3 git \
&& rm -rf /var/lib/apt/lists/*
RUN git config --system --add safe.directory '*'
WORKDIR /app
COPY ./target/release/git-hook /bin
EXPOSE 8083
CMD ["git-hook"]

9
docker/gitserver.Dockerfile
View File

@ -1,9 +0,0 @@
FROM ubuntu:24.04
RUN apt-get update && apt-get install -y --no-install-recommends \
ca-certificates libssl3 git openssh-client \
&& rm -rf /var/lib/apt/lists/*
RUN git config --system --add safe.directory '*'
WORKDIR /app
COPY ./target/release/gitserver /bin
EXPOSE 8021 2222
CMD ["gitserver"]

8
docker/metrics.Dockerfile
View File

@ -1,8 +0,0 @@
FROM ubuntu:24.04
RUN apt-get update && apt-get install -y --no-install-recommends \
ca-certificates libssl3 \
&& rm -rf /var/lib/apt/lists/*
WORKDIR /app
COPY ./target/release/metrics-aggregator /bin
EXPOSE 9090
CMD ["metrics-aggregator"]

8
docker/static.Dockerfile
View File

@ -1,8 +0,0 @@
FROM ubuntu:24.04
RUN apt-get update && apt-get install -y --no-install-recommends \
ca-certificates libssl3 \
&& rm -rf /var/lib/apt/lists/*
WORKDIR /app
COPY ./target/release/static-server /bin
EXPOSE 8081
CMD ["static-server"]

8
eslint.config.js
View File

@ -16,15 +16,7 @@ export default defineConfig([
reactRefresh.configs.vite, reactRefresh.configs.vite,
], ],
languageOptions: { languageOptions: {
ecmaVersion: 2020,
globals: globals.browser, globals: globals.browser,
}, },
rules: {
'react-refresh/only-export-components': [
'warn',
{ allowExportNames: ['useThemeCustomization', 'useThemePreset', 'useRoom', 'useOptionalRoom', 'resetAllThemeVars', 'loadThemeVars', 'applyThemePreset'] },
],
'react-hooks/exhaustive-deps': 'warn',
},
}, },
]) ])

901
gene.md
View File

@ -1,901 +0,0 @@
# Gene 方案
## 这份文档针对什么
这个项目里，`Skill` 已经不是一个抽象概念，而是完整的业务实体：
* 后端有 `project_skill` 持久化模型
* Git 同步会扫描仓库里的 `SKILL.md`
* 聊天构建会把启用的技能注入上下文
* 前端有技能列表、详情、编辑、删除、扫描
* 内建技能模板也已经存在
所以这里不应该“用 Gene 替换 Skill”，而应该是：
* `Skill` 继续负责执行和交付能力
* `Gene` 作为新增元层，负责让技能可演化、可比较、可追踪、可继承
---
## 设计结论
`Gene` 的正确位置不是新的执行系统，也不是 `Skill` 的替代品，而是 `Skill` 的生命周期治理层。
```text
Skill = 可执行内容 + 上下文注入 + 业务交付
Gene = 演化族 + 版本谱系 + 评估记录 + 选择记录
```
更准确地说：
```text
Skill 是可执行的技能内容，负责被扫描、编辑、启用、注入和交付能力。
Gene 是 Skill 的演化族，负责组织一个 Skill 在项目内的版本、来源、谱系、评估和选择记录。
GeneRevision 是 Gene 下的不可变版本节点，绑定到某个 Skill 内容快照。
GeneEvaluation 是对某个 GeneRevision 的评估结果。
GeneSelection 是对当前有效 GeneRevision 的显式选择记录。
```
---
## 项目现状
### Skill 已经落在这些地方
* `libs/models/projects/project_skill.rs`
* 定义了 `project_skill` 实体
* 已包含 `source`
* 已包含 `repo_id`
* 已包含 `commit_sha`
* 已包含 `blob_hash`
* 已包含 `content`
* 已包含 `metadata`
* 已包含 `enabled`
* `libs/git/hook/sync/mod.rs`
* 扫描仓库中的 `SKILL.md`
* 从 frontmatter 解析 `name`、`description`、`license`、`compatibility`
* 用 `commit_sha` 和 `blob_hash` 做增量同步
* `libs/agent/skills/templates.rs`
* 内建技能模板已经编译进程序
* 这些模板本质上是“系统内置 Skill”
* `libs/agent/chat/message_builder.rs`
* 读取项目中启用的技能
* 把技能注入对话上下文
* 和 embedding / perception 一起影响模型行为
* `src/app/project/skills/*`
* 已有技能管理 UI
* 能查看、编辑、删除、扫描技能
---
## 现有 Skill 的问题
当前 `Skill` 已经能用，但还不够“可进化”：
* 只有内容，没有明确的演化关系
* 只有当前状态，没有谱系
* 只有启用/禁用，没有版本选择
* 只有来源信息，没有变体比较
* 只有同步时间和 blob hash，没有“为什么变成这样”的记录
* 有评估空间，但没有评估结果和版本绑定
* 有扫描和编辑能力，但没有可审计的回滚与选择记录
换句话说，现在的 `Skill` 更像“静态资产”，还不是“进化单元”。
---
## 设计原则
### 1. Skill 仍然是唯一执行实体
`Skill` 继续负责：
* 被扫描
* 被编辑
* 被启用或禁用
* 被注入聊天上下文
* 交付实际能力
`Gene` 不直接执行任务，不直接调用工具，不直接参与上下文拼装。
---
### 2. Gene 只管理 Skill 的演化元数据
`Gene` 管理的是：
* 版本
* 来源
* 父子关系
* 变体
* 评估
* 选择
* 淘汰
* 回滚
它不应该成为第二套 `Skill` 系统。
---
### 3. GeneRevision 必须不可变
`Skill` 可以是当前可编辑对象。
`GeneRevision` 是不可变演化节点。一旦创建，不应再修改：
* 内容快照
* 父代关系
* 来源信息
* `commit_sha`
* `blob_hash`
* `content_hash`
* `mutation_reason`
* `mutation_diff`
后续任何内容变化、prompt 变化、工具权限变化、上下文注入规则变化，都应该产生新的 `GeneRevision`。
---
### 4. Evaluation 必须绑定到具体 Revision
评估不是评价一个抽象的 `Gene`，而是评价某个具体版本。
因此：
```text
GeneEvaluation 必须绑定 revision_id。
```
否则同一个 `Gene` 下存在多个版本时，评估结果无法精确归因。
---
### 5. Selection 必须显式记录
如果系统选择了某个版本作为当前有效版本，必须记录：
* 选中了哪个 revision
* 为什么选它
* 谁选的
* 依据什么策略选的
* 什么时候选的
* 是否仍然 active
这能支持审计、回滚和后续自动选择。
---
### 6. SKILL.md 仍然是仓库内 Skill 的事实来源
仓库里的 `SKILL.md` 仍然是 Git 同步的事实来源。
`Gene` 只记录它如何演化，不改变仓库同步的基本语义。
---
## 核心概念
### Skill
`Skill` 是现有业务实体。
它负责：
```text
可执行内容
上下文注入
用户可见编辑
Git 扫描
启用 / 禁用
业务交付
```
### Gene
`Gene` 是某个 Skill 的演化族。
它负责组织这个 Skill 的生命周期：
```text
这个能力从哪里来
经历过哪些版本
有哪些变体
评估结果如何
当前选择哪个版本
哪些版本被淘汰
```
### GeneRevision
`GeneRevision` 是 `Gene` 下的一个不可变版本节点。
它绑定某个 `Skill` 的内容状态，例如：
```text
skill_id
skill_slug
commit_sha
blob_hash
content_hash
content_snapshot_ref
```
### GeneEvaluation
`GeneEvaluation` 是对某个 `GeneRevision` 的评估结果。
它回答：
```text
这个版本好不好？
在哪个数据集上测的？
指标是什么？
是否通过？
成本和延迟如何？
失败样本是什么？
```
### GeneSelection
`GeneSelection` 是对当前有效版本的显式选择记录。
它回答：
```text
当前选中哪个 revision？
为什么选它？
谁选的？
依据什么策略？
是否还在生效？
```
---
## Gene 和 Skill 的关系
可以把关系理解为：
```text
Gene 1 ── has many ── GeneRevision
GeneRevision ── references ── Skill snapshot
GeneRevision ── has many ── GeneEvaluation
Gene ── has one active ── GeneSelection
GeneSelection ── selects ── GeneRevision
Skill ── executes ── actual capability
```
也就是说：
* `Skill` 解决“能不能做”
* `Gene` 解决“该保留哪个版本、为什么保留、怎么变体、怎么传播”
* `GeneRevision` 解决“这个能力在某一刻具体长什么样”
* `GeneEvaluation` 解决“这个版本是否足够好”
* `GeneSelection` 解决“当前应该用哪个版本”
---
## 数据模型
### ProjectGene
```text
ProjectGene
- gene_id
- project_uuid
- skill_id
- skill_slug
- name
- description
- owner
- status
- created_at
- updated_at
```
说明：
* `gene_id` 是 Gene 的唯一标识
* `project_uuid` 绑定项目
* `skill_id` / `skill_slug` 绑定现有 Skill
* `status` 可为 `active`、`archived`、`deprecated`
* `owner` 用于责任归属
---
### ProjectGeneRevision
```text
ProjectGeneRevision
- revision_id
- gene_id
- version
- parent_revision_id
- origin
- source
- skill_id
- skill_slug
- commit_sha
- blob_hash
- content_hash
- content_snapshot_ref
- mutation_reason
- mutation_diff
- created_by
- created_at
```
说明：
* `revision_id` 是内部唯一版本节点
* `version` 是用户可见版本号，不承担唯一性职责
* `parent_revision_id` 表示单父版本关系
* `origin` 表示来源，例如 `git_sync`、`manual_edit`、`builtin_template`、`migration`
* `source` 复用现有 `Skill.source`
* `commit_sha` / `blob_hash` 记录 Git 来源
* `content_hash` 记录内容稳定标识
* `content_snapshot_ref` 用于指向当时的内容快照
* `mutation_reason` 记录为什么产生这个版本
* `mutation_diff` 记录相对父版本的变化
---
### ProjectGeneEvaluation
```text
ProjectGeneEvaluation
- evaluation_id
- gene_id
- revision_id
- eval_name
- evaluator
- dataset_ref
- dataset_version
- metric_name
- score
- threshold
- passed
- sample_count
- failure_count
- latency_ms_avg
- cost
- result_summary
- result_json
- evaluated_at
```
说明：
* `revision_id` 必填
* `score` 不应脱离 `metric_name` 单独解释
* `threshold` 用于判断是否通过
* `sample_count` 和 `failure_count` 用于判断评估可信度
* `result_json` 存放详细结果、失败样本、分项指标等
---
### ProjectGeneSelection
```text
ProjectGeneSelection
- selection_id
- gene_id
- selected_revision_id
- project_uuid
- policy
- reason
- selected_by
- selected_at
- active
```
说明：
* 同一个 `gene_id` 同一时间只能有一个 active selection
* `policy` 可以是 `manual`、`latest_passed`、`best_score`、`stable_low_cost`
* `reason` 用于审计和回滚
* `selected_by` 可以是用户、系统或自动策略
---
## 关于 GeneLineage
MVP 阶段不单独引入 `GeneLineage` 表。
原因是：如果每个版本只有一个父版本，`ProjectGeneRevision.parent_revision_id` 已经足够表达谱系。
```text
MVP：单父版本树
Future：多父 DAG / merge / cross-skill inheritance
```
未来如果需要支持合并、交叉继承或复杂谱系，再引入：
```text
ProjectGeneEdge
- parent_revision_id
- child_revision_id
- edge_type
- mutation_reason
- mutation_diff
```
---
## 关于 Variant
`Variant` 是 Gene 的重要能力，但不一定是 MVP 的独立表。
这些变化都可以先作为新的 `GeneRevision` 表达：
* 更严格的 prompt
* 更短的 prompt
* 不同工具权限
* 不同上下文注入规则
* 不同评估阈值
MVP 中：
```text
不单独引入 ProjectGeneVariant。
所有变体先作为 GeneRevision 表达。
```
当系统需要以下能力时，再引入 `GeneExperiment` / `GeneVariant`：
* A/B 实验
* 并行流量分配
* 实验分组
* 统计显著性
* 多候选版本同时比较
---
## Git 同步流程
当前 Git 同步已经会扫描仓库中的 `SKILL.md`，并使用 `commit_sha` 和 `blob_hash` 做增量同步。
引入 Gene 后，Git 同步流程建议变成：
```text
当 Git 同步发现 SKILL.md 的 blob_hash 变化时：
1. 正常创建或更新 project_skill。
2. 查找该 skill 对应的 project_gene。
3. 如果不存在 project_gene，则创建一个。
4. 查找该 gene 下最新的 project_gene_revision。
5. 如果新的 blob_hash / content_hash 不同，则创建新的 GeneRevision。
6. 将上一 revision 设为 parent_revision_id。
7. mutation_reason 默认记录为 git_sync。
8. mutation_diff 记录上一个 SKILL.md 与当前 SKILL.md 的 diff。
9. 不自动改变 selected_revision，除非选择策略明确允许。
```
关键约束：
```text
Git sync 可以产生新的 GeneRevision。
Git sync 不应默认切换当前线上选中版本。
```
这样可以避免仓库变更自动导致线上行为漂移。
---
## 手动编辑流程
当用户在 UI 中编辑 `Skill` 内容时：
```text
1. 更新 project_skill。
2. 计算新的 content_hash。
3. 查找对应 project_gene。
4. 创建新的 project_gene_revision。
5. parent_revision_id 指向编辑前的 revision。
6. mutation_reason 记录为 manual_edit。
7. mutation_diff 记录编辑前后的差异。
8. 可选择是否自动把新 revision 标记为 selected。
```
建议 MVP 默认：
```text
手动编辑后创建新 revision，但不自动覆盖 active selection。
由用户显式选择是否启用该 revision。
```
如果产品希望“编辑即生效”，也可以让 selection 同步更新，但必须记录：
```text
policy = manual_edit_auto_select
reason = "User edited skill content"
```
---
## 聊天上下文注入流程
现有流程中，`libs/agent/chat/message_builder.rs` 读取项目中启用的 `Skill`，并把技能注入对话上下文。
引入 Gene 后，这个原则不变：
```text
message_builder 不直接读取 Gene 内容。
message_builder 仍然读取启用的 Skill。
Gene 只影响哪个 Skill revision 被视为推荐版本或当前选中版本。
```
如果未来要让 `GeneSelection` 影响上下文注入，推荐路径是：
```text
GeneSelection
-> resolve selected GeneRevision
-> materialize / update project_skill
-> message_builder reads project_skill
```
不建议：
```text
message_builder
-> read Gene
-> assemble prompt
```
因为这会让 `Gene` 偷偷变成新的执行层。
---
## API 设计
MVP API 可以包括：
```text
GET /projects/:project_uuid/skills/:skill_id/gene
POST /projects/:project_uuid/skills/:skill_id/gene
GET /projects/:project_uuid/genes/:gene_id/revisions
POST /projects/:project_uuid/genes/:gene_id/revisions
GET /projects/:project_uuid/genes/:gene_id/evaluations
POST /projects/:project_uuid/genes/:gene_id/evaluations
GET /projects/:project_uuid/genes/:gene_id/selection
POST /projects/:project_uuid/genes/:gene_id/select
```
选择接口示例：
```json
{
"selected_revision_id": "rev_123",
"policy": "manual",
"reason": "Higher pass rate on regression eval"
}
```
评估写入接口示例：
```json
{
"revision_id": "rev_123",
"eval_name": "skill_regression_eval",
"dataset_ref": "datasets/skill-regression-v1",
"dataset_version": "2026-01-01",
"metric_name": "task_success_rate",
"score": 0.92,
"threshold": 0.85,
"passed": true,
"sample_count": 100,
"failure_count": 8,
"latency_ms_avg": 1200,
"cost": 0.34
}
```
---
## UI 设计
`Gene` 不替代现有技能管理 UI。
推荐把它放在 `Skill Detail` 页面中的一个“演化”标签页。
```text
Skill Detail
- 基本信息
- 内容编辑
- 启用状态
- Evolution / Gene
- 当前选中 revision
- 版本列表
- 父子关系
- 每个版本的 diff
- 每个版本的评估结果
- 选择按钮
- 回滚按钮
```
MVP UI 可以先做只读：
```text
1. 显示 Gene 信息
2. 显示 Revision 列表
3. 显示每个 Revision 的来源、时间、commit_sha、blob_hash
4. 显示相邻 Revision 的 diff
```
第二阶段再加入：
```text
1. 手动选择 revision
2. 回滚 revision
3. 展示评估结果
4. 根据评估结果推荐版本
```
---
## 评估指标
Gene 的核心不是“更像生物学”，而是“更像可验证的演化对象”。
每个 `GeneRevision` 都应该支持评估，例如：
* 任务成功率
* 失败率
* 平均耗时
* 误触发率
* 人工接受率
* 回滚率
* 成本
* 稳定性
* 安全失败数
* 用户满意度
没有评估的 Gene，只是重命名后的 Skill 管理。
---
## 选择策略
如果存在多个 GeneRevision，系统应该能选择更优版本。
MVP 阶段不做自动选择，只做人工选择和可审计回滚。
后续选择规则可以逐步加入：
```text
1. 先看是否通过基础测试
2. 再看近期成功率
3. 再看失败率
4. 再看成本
5. 再看延迟
6. 再看稳定性
7. 再看人工接受率
```
可选策略：
```text
manual
latest_passed
best_score
stable_low_cost
lowest_latency
highest_acceptance_rate
```
自动选择必须满足：
```text
有评估数据
有样本量
有失败分类
有回滚机制
有选择审计记录
```
---
## Migration 策略
对于已有的 `project_skill`，可以执行一次初始化迁移：
```text
对每个 project_skill：
1. 创建 project_gene。
2. 创建初始 project_gene_revision。
3. revision.origin = migration。
4. revision.skill_id = project_skill.id。
5. revision.skill_slug = project_skill.slug。
6. revision.commit_sha = project_skill.commit_sha。
7. revision.blob_hash = project_skill.blob_hash。
8. revision.content_hash = hash(project_skill.content)。
9. revision.content_snapshot_ref = 当前 Skill 内容快照。
10. 创建 active project_gene_selection。
11. selected_revision_id = 初始 revision。
12. policy = migration。
13. reason = "Initial gene selection from existing project_skill"。
```
这样现有 Skill 都可以无损进入 Gene 生命周期模型。
---
## 推荐实现顺序
```text
1. 保持 Skill 执行、扫描、编辑、注入流程不变。
2. 增加 project_gene 和 project_gene_revision 表。
3. 为现有 project_skill 执行一次 migration：每个 Skill 创建一个 Gene 和初始 Revision。
4. 在 Git sync 发现 blob_hash 变化时，自动创建新的 GeneRevision。
5. 在 Skill 详情页增加只读版本历史和 diff 展示。
6. 增加 ProjectGeneEvaluation 表和手动写入 API。
7. 增加 ProjectGeneSelection 表和人工选择 / 回滚能力。
8. 当评估数据稳定后，再做自动选择策略。
9. 最后再考虑 Variant / Experiment / A-B 测试。
```
---
## 落地判断标准
一个能力如果只是：
* 能被扫描
* 能被编辑
* 能被启用或禁用
* 能被注入上下文
* 能交付业务能力
它还是 `Skill`。
一个能力如果还能：
* 被追踪来源
* 被比较版本
* 被记录变体
* 被评估好坏
* 被继承和淘汰
* 被审计选择
* 被安全回滚
它才进入 `Gene` 管理范畴。
---
## 非目标
Gene MVP 不做以下事情：
```text
1. 不替换 project_skill。
2. 不改变 SKILL.md 作为仓库技能事实来源的地位。
3. 不直接参与聊天上下文构建。
4. 不直接执行工具调用。
5. 不自动改写 Skill 内容。
6. 不在没有评估和回滚机制的情况下自动切换线上版本。
7. 不一开始支持复杂遗传算法、交叉、随机变异或自动进化。
8. 不新增一套和 Skill 并列的执行 UI。
```
---
## 风险与约束
### 1. Gene 变成另一个 Skill
风险：
```text
Gene 中也开始存 prompt、工具权限、上下文注入规则，并且聊天时直接读取 Gene。
```
规避：
```text
Gene 不存放执行主内容。
执行内容仍然归 Skill 所有。
GeneRevision 只引用或快照 Skill 的某个状态。
```
---
### 2. 评估绑定到可变 Skill
风险：
```text
评估结果只挂在 skill_id 上，Skill 内容变化后，评估记录失真。
```
规避：
```text
评估必须绑定 revision_id + content_hash / blob_hash。
```
---
### 3. 自动选择过早上线
风险：
```text
没有足够评估数据时自动切换版本，导致线上行为漂移。
```
规避：
```text
MVP 只做人工选择和可审计回滚。
自动选择必须依赖稳定评估和回滚机制。
```
---
### 4. mutation_diff 膨胀
风险：
```text
每个版本都保存完整 diff，长期可能膨胀。
```
规避：
```text
MVP 可直接存 mutation_diff。
后续引入 mutation_diff_ref 或对象存储引用。
```
---
### 5. version 语义不清
风险：
```text
version 同时承担用户展示、唯一标识、Git 版本等多种含义。
```
规避：
```text
revision_id = 系统内部唯一版本节点
version = 用户可见版本号
commit_sha / blob_hash = Git 来源版本标识
content_hash = 内容稳定标识
```
---
## 最终结论
这个项目已经具备 `Skill` 的完整工程闭环。
`Gene` 的正确位置不是替换它，而是补上它缺失的演化层：
```text
Skill 负责落地执行
Gene 负责生命周期治理
GeneRevision 负责不可变版本节点
GeneEvaluation 负责版本质量判断
GeneSelection 负责显式选择和回滚
```
两者结合后，技能系统才从“可配置”变成“可进化”。

9
index.html
View File

@ -2,13 +2,12 @@
<html lang="en"> <html lang="en">
<head> <head>
<meta charset="UTF-8" /> <meta charset="UTF-8" />
<link href="/logo.png" rel="icon" type="image/svg+xml"/> <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
<meta content="width=device-width, initial-scale=1.0" name="viewport"/> <meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>GitData.AI</title> <title>GitDataAI</title>
<link rel="preload" href="/@fs/node_modules/@fontsource-variable/geist/files/Geist%5Bwght%5D.woff2" as="font" type="font/woff2" crossorigin />
</head> </head>
<body> <body>
<div id="root"></div> <div id="root"></div>
<script src="/src/main.tsx" type="module"></script> <script type="module" src="/src/main.tsx"></script>
</body> </body>
</html> </html>

1
lib.rs
View File

@ -1 +0,0 @@
// Frontend embedding is handled by libs/frontend crate. ci

47
libs/agent/Cargo.toml
View File

@ -1,47 +0,0 @@
[package]
name = "agent"
version.workspace = true
edition.workspace = true
authors.workspace = true
description.workspace = true
repository.workspace = true
readme.workspace = true
homepage.workspace = true
license.workspace = true
keywords.workspace = true
categories.workspace = true
documentation.workspace = true
[lib]
path = "lib.rs"
name = "agent"
[features]
default = ["rig"]
rig = []
[dependencies]
rig-core = { workspace = true, features = ["derive"] }
tokio = { workspace = true }
async-trait = { workspace = true }
qdrant-client = { workspace = true }
sea-orm = { workspace = true }
serde = { workspace = true, features = ["derive"] }
serde_json = { workspace = true }
thiserror = { workspace = true }
db = { workspace = true }
config = { path = "../config" }
models = { workspace = true }
chrono = { workspace = true }
uuid = { workspace = true, features = ["v7"] }
futures = { workspace = true }
tiktoken-rs = { workspace = true }
once_cell = { workspace = true }
regex = { workspace = true }
tracing = { workspace = true }
metrics = { workspace = true }
rust_decimal = { workspace = true }
reqwest = { workspace = true, features = ["json"] }
utoipa = { workspace = true }
tokio-stream = { workspace = true }
redis = { workspace = true, features = ["tokio-comp"] }
queue = { workspace = true }
[lints]
workspace = true

4
libs/agent/agent/mod.rs
View File

@ -1,4 +0,0 @@
//! Rig-based agent using rig's built-in Agent with full feature support.
pub mod rig_tool;
pub use rig_tool::{AgentResponse, RigAgentService, StreamChunk};

234
libs/agent/agent/rig_tool.rs
View File

@ -1,234 +0,0 @@
use futures::Stream;
use futures::StreamExt;
use rig::{
agent::{AgentBuilder, MultiTurnStreamItem},
client::CompletionClient,
completion::Prompt,
streaming::{StreamedAssistantContent, StreamingPrompt},
};
use tokio::sync::mpsc;
use tokio_stream::wrappers::ReceiverStream;
use crate::client::AiClientConfig;
use crate::error::AgentError;
#[derive(Debug)]
pub struct AgentResponse {
pub content: String,
pub input_tokens: u64,
pub output_tokens: u64,
}
#[derive(Debug)]
pub enum StreamChunk {
Text(String),
Final {
content: String,
input_tokens: u64,
output_tokens: u64,
},
}
pub struct RigAgentService {
config: AiClientConfig,
model_name: String,
}
impl RigAgentService {
pub fn new(config: AiClientConfig, model_name: impl Into<String>) -> Self {
Self {
config,
model_name: model_name.into(),
}
}
pub async fn prompt(
&self,
system_prompt: &str,
user_input: &str,
) -> std::result::Result<AgentResponse, AgentError> {
let client = self.config.build_rig_client();
let model = client.completion_model(&self.model_name);
let agent = AgentBuilder::new(model).preamble(system_prompt).build();
let response = agent
.prompt(user_input)
.extended_details()
.await
.map_err(|e: rig::completion::PromptError| AgentError::OpenAi(e.to_string()))?;
Ok(AgentResponse {
content: response.output,
input_tokens: response.usage.input_tokens,
output_tokens: response.usage.output_tokens,
})
}
pub async fn prompt_with_tools(
&self,
system_prompt: &str,
user_input: &str,
tools: Vec<Box<dyn rig::tool::ToolDyn + 'static>>,
max_turns: usize,
) -> std::result::Result<AgentResponse, AgentError> {
let client = self.config.build_rig_client();
let model = client.completion_model(&self.model_name);
let agent = AgentBuilder::new(model)
.preamble(system_prompt)
.tools(tools)
.default_max_turns(max_turns)
.build();
let response = agent
.prompt(user_input)
.max_turns(max_turns)
.extended_details()
.await
.map_err(|e: rig::completion::PromptError| AgentError::OpenAi(e.to_string()))?;
Ok(AgentResponse {
content: response.output,
input_tokens: response.usage.input_tokens,
output_tokens: response.usage.output_tokens,
})
}
pub async fn stream_prompt(
&self,
system_prompt: &str,
user_input: &str,
) -> std::result::Result<
impl Stream<Item = std::result::Result<StreamChunk, AgentError>>,
AgentError,
> {
let client = self.config.build_rig_client();
let model = client.completion_model(&self.model_name);
let agent = AgentBuilder::new(model).preamble(system_prompt).build();
let stream: rig::agent::StreamingResult<_> = agent.stream_prompt(user_input).await;
let (tx, rx) = mpsc::channel::<std::result::Result<StreamChunk, AgentError>>(100);
tokio::spawn(async move {
let mut final_content = String::new();
tokio::pin!(stream);
while let Some(item) = stream.next().await {
match item {
Ok(MultiTurnStreamItem::StreamAssistantItem(
StreamedAssistantContent::Text(text),
)) => {
let _ = tx.send(Ok(StreamChunk::Text(text.text.clone()))).await;
final_content.push_str(&text.text);
}
Ok(MultiTurnStreamItem::StreamAssistantItem(
StreamedAssistantContent::ToolCall {
tool_call,
internal_call_id: _,
},
)) => {
let args_str = match &tool_call.function.arguments {
serde_json::Value::String(s) => s.clone(),
v => serde_json::to_string(v).unwrap_or_default(),
};
tracing::info!(
tool = %tool_call.function.name,
args = %args_str,
"rig_agent_streaming_tool_call"
);
}
Ok(MultiTurnStreamItem::StreamUserItem(
rig::streaming::StreamedUserContent::ToolResult { tool_result, .. },
)) => {
tracing::info!(
tool_result_id = %tool_result.id,
"rig_agent_streaming_tool_result"
);
}
Ok(MultiTurnStreamItem::FinalResponse(resp)) => {
let usage = resp.usage();
let _ = tx
.send(Ok(StreamChunk::Final {
content: final_content.clone(),
input_tokens: usage.input_tokens,
output_tokens: usage.output_tokens,
}))
.await;
}
Err(e) => {
let _ = tx.send(Err(AgentError::OpenAi(e.to_string()))).await;
}
_ => {}
}
}
});
Ok(ReceiverStream::new(rx))
}
pub async fn stream_prompt_with_tools(
&self,
system_prompt: &str,
user_input: &str,
tools: Vec<Box<dyn rig::tool::ToolDyn + 'static>>,
max_turns: usize,
) -> std::result::Result<
impl Stream<Item = std::result::Result<StreamChunk, AgentError>>,
AgentError,
> {
let client = self.config.build_rig_client();
let model = client.completion_model(&self.model_name);
let agent = AgentBuilder::new(model)
.preamble(system_prompt)
.tools(tools)
.default_max_turns(max_turns)
.build();
let stream = agent
.stream_prompt(user_input)
.with_history(Vec::<rig::completion::Message>::new())
.multi_turn(max_turns)
.await;
let (tx, rx) = mpsc::channel::<Result<StreamChunk, AgentError>>(100);
tokio::spawn(async move {
let mut final_content = String::new();
tokio::pin!(stream);
while let Some(item) = stream.next().await {
match item {
Ok(MultiTurnStreamItem::StreamAssistantItem(
StreamedAssistantContent::Text(text),
)) => {
let _ = tx.send(Ok(StreamChunk::Text(text.text.clone()))).await;
final_content.push_str(&text.text);
}
Ok(MultiTurnStreamItem::FinalResponse(resp)) => {
let usage = resp.usage();
let _ = tx
.send(Ok(StreamChunk::Final {
content: final_content.clone(),
input_tokens: usage.input_tokens,
output_tokens: usage.output_tokens,
}))
.await;
}
Err(e) => {
let _ = tx.send(Err(AgentError::OpenAi(e.to_string()))).await;
}
_ => {}
}
}
});
Ok(ReceiverStream::new(rx))
}
pub fn count_tokens(&self, text: &str) -> Result<usize, AgentError> {
crate::tokent::count_text(text, &self.model_name)
.map_err(|e| AgentError::Internal(e.to_string()))
}
}

668
libs/agent/billing.rs
View File

@ -1,668 +0,0 @@
//! Billing service — handles user-level and project-level billing, deduction,
//! credit initialization, and error persistence.
//!
//! Architecture:
//! - Each user gets $10 personal balance on signup.
//! - Each project gets $20 balance only if it's the creator's first project,
//! $0 otherwise.
//! - AI usage is deducted from the project balance first; if insufficient,
//! falls through to the user's personal balance.
//! - Monthly quota only applies to pro users (is_pro = true).
//! - If both project and user balance are insufficient, a billing_error
//! record is persisted and an error is returned to the caller.
use db::database::AppDatabase;
use models::agents::model_pricing;
use models::ai::billing_error;
use models::projects::{project, project_billing, project_billing_history};
use models::users::{user_billing, user_billing_history};
use rust_decimal::Decimal;
use sea_orm::*;
use uuid::Uuid;
use crate::error::AgentError;
fn default_user_balance() -> Decimal {
Decimal::new(100_000, 4)
} // $10.0000
fn first_project_credit() -> Decimal {
Decimal::new(200_000, 4)
} // $20.0000
const SUBSEQUENT_PROJECT_BALANCE: Decimal = Decimal::ZERO;
#[derive(Debug, Clone, serde::Serialize, serde::Deserialize, utoipa::ToSchema)]
pub struct BillingRecord {
pub cost: f64,
pub currency: String,
pub input_tokens: i64,
pub output_tokens: i64,
pub deducted_from: String, // "project" or "user"
}
#[derive(Debug)]
pub enum BillingResult {
Success(BillingRecord),
InsufficientBalance { message: String },
}
/// Record AI usage: deduct from project balance first, fall through to user balance.
///
/// Returns `InsufficientBalance` if neither account can cover the cost.
/// On insufficient balance, a `billing_error` record is persisted for frontend display.
pub async fn record_ai_usage(
db: &AppDatabase,
project_uid: Uuid,
user_uid: Uuid,
model_id: Uuid,
input_tokens: i64,
output_tokens: i64,
) -> Result<BillingResult, AgentError> {
let total_cost = compute_cost(db, model_id, input_tokens, output_tokens).await?;
let currency = get_currency(db, model_id).await?;
// Verify project exists
let _ = project::Entity::find_by_id(project_uid)
.one(db)
.await?
.ok_or_else(|| AgentError::Internal("Project not found".into()))?;
// Attempt project-level deduction first
let project_result = deduct_from_project(
db,
project_uid,
total_cost,
&currency,
model_id,
input_tokens,
output_tokens,
)
.await;
match project_result {
Ok(()) => {
let cost_f64 = decimal_to_f64(total_cost);
tracing::info!(
project_id = %project_uid,
model_id = %model_id,
input_tokens, output_tokens,
cost = %cost_f64,
currency = %currency,
deducted_from = "project",
"ai_usage_recorded"
);
Ok(BillingResult::Success(BillingRecord {
cost: cost_f64,
currency,
input_tokens,
output_tokens,
deducted_from: "project".to_string(),
}))
}
Err(_) => {
// Project balance insufficient — try user personal balance
let user_result = deduct_from_user(
db,
user_uid,
total_cost,
&currency,
project_uid,
model_id,
input_tokens,
output_tokens,
)
.await;
match user_result {
Ok(()) => {
let cost_f64 = decimal_to_f64(total_cost);
tracing::info!(
user_id = %user_uid,
project_id = %project_uid,
model_id = %model_id,
input_tokens, output_tokens,
cost = %cost_f64,
currency = %currency,
deducted_from = "user",
"ai_usage_recorded"
);
Ok(BillingResult::Success(BillingRecord {
cost: cost_f64,
currency,
input_tokens,
output_tokens,
deducted_from: "user".to_string(),
}))
}
Err(insufficient_msg) => {
// Both project and user balance insufficient — persist error
persist_billing_error(
db,
"project",
project_uid,
"insufficient_balance",
&insufficient_msg,
Some(serde_json::json!({
"user_id": user_uid.to_string(),
"model_id": model_id.to_string(),
"input_tokens": input_tokens,
"output_tokens": output_tokens,
"cost": decimal_to_f64(total_cost),
"currency": currency,
})),
)
.await?;
Ok(BillingResult::InsufficientBalance {
message: insufficient_msg,
})
}
}
}
}
}
/// Record personal AI usage against the user's own billing balance.
pub async fn record_user_ai_usage(
db: &AppDatabase,
user_uid: Uuid,
model_id: Uuid,
input_tokens: i64,
output_tokens: i64,
) -> Result<BillingResult, AgentError> {
let total_cost = compute_cost(db, model_id, input_tokens, output_tokens).await?;
let currency = get_currency(db, model_id).await?;
match deduct_from_user_personal(
db,
user_uid,
total_cost,
&currency,
model_id,
input_tokens,
output_tokens,
)
.await
{
Ok(()) => {
let cost_f64 = decimal_to_f64(total_cost);
tracing::info!(
user_id = %user_uid,
model_id = %model_id,
input_tokens, output_tokens,
cost = %cost_f64,
currency = %currency,
deducted_from = "user",
scope = "personal",
"ai_usage_recorded"
);
Ok(BillingResult::Success(BillingRecord {
cost: cost_f64,
currency,
input_tokens,
output_tokens,
deducted_from: "user".to_string(),
}))
}
Err(insufficient_msg) => {
persist_billing_error(
db,
"user",
user_uid,
"insufficient_balance",
&insufficient_msg,
Some(serde_json::json!({
"user_id": user_uid.to_string(),
"model_id": model_id.to_string(),
"input_tokens": input_tokens,
"output_tokens": output_tokens,
"cost": decimal_to_f64(total_cost),
"currency": currency,
"scope": "personal",
})),
)
.await?;
Ok(BillingResult::InsufficientBalance {
message: insufficient_msg,
})
}
}
}
/// Check whether a project + user has sufficient combined balance for a potential AI call.
/// Called before starting AI processing to avoid wasted compute.
pub async fn check_balance(
db: &AppDatabase,
project_uid: Uuid,
user_uid: Uuid,
model_id: Uuid,
estimated_input_tokens: i64,
estimated_output_tokens: i64,
) -> Result<bool, AgentError> {
let estimated_cost = compute_cost(
db,
model_id,
estimated_input_tokens,
estimated_output_tokens,
)
.await?;
let project_balance = get_project_balance(db, project_uid).await;
let user_balance = get_user_balance(db, user_uid).await;
Ok(project_balance + user_balance >= estimated_cost)
}
/// Check whether a user's personal balance can cover a potential AI call.
pub async fn check_user_balance(
db: &AppDatabase,
user_uid: Uuid,
model_id: Uuid,
estimated_input_tokens: i64,
estimated_output_tokens: i64,
) -> Result<bool, AgentError> {
let estimated_cost = compute_cost(
db,
model_id,
estimated_input_tokens,
estimated_output_tokens,
)
.await?;
let user_balance = get_user_balance(db, user_uid).await;
Ok(user_balance >= estimated_cost)
}
// ── Initialization ──
/// Initialize a user billing account with the default $10 balance.
/// Called on user signup / first login.
pub async fn initialize_user_billing(db: &AppDatabase, user_uid: Uuid) -> Result<(), AgentError> {
let now = chrono::Utc::now();
user_billing::ActiveModel {
user: Set(user_uid),
balance: Set(default_user_balance()),
currency: Set("USD".to_string()),
is_pro: Set(false),
monthly_quota: Set(Decimal::ZERO),
month_used: Set(Decimal::ZERO),
cycle_start: Set(None),
cycle_end: Set(None),
updated_at: Set(now),
created_at: Set(now),
}
.insert(db)
.await
.map_err(|e| AgentError::Internal(format!("failed to create user billing: {}", e)))?;
tracing::info!(user_id = %user_uid, balance = "$10", "user_billing_initialized");
Ok(())
}
/// Initialize a project billing account.
/// Grants $20 only if this is the creator's first project; $0 otherwise.
pub async fn initialize_project_billing(
db: &AppDatabase,
project_uid: Uuid,
creator_uid: Uuid,
) -> Result<(), AgentError> {
// Check how many projects this user has already created
let existing_count = project::Entity::find()
.filter(project::Column::CreatedBy.eq(creator_uid))
.filter(project::Column::Id.ne(project_uid))
.count(db)
.await
.map_err(|e| AgentError::Internal(format!("failed to count user projects: {}", e)))?;
let is_first = existing_count == 0;
let initial_balance = if is_first {
first_project_credit()
} else {
SUBSEQUENT_PROJECT_BALANCE
};
let now = chrono::Utc::now();
project_billing::ActiveModel {
project: Set(project_uid),
balance: Set(initial_balance),
currency: Set("USD".to_string()),
user: Set(Some(creator_uid)),
initial_credit_granted: Set(is_first),
is_pro: Set(false),
monthly_quota: Set(Decimal::ZERO),
month_used: Set(Decimal::ZERO),
cycle_start: Set(None),
cycle_end: Set(None),
updated_at: Set(now),
created_at: Set(now),
}
.insert(db)
.await
.map_err(|e| AgentError::Internal(format!("failed to create project billing: {}", e)))?;
if is_first {
// Record the credit in billing history
project_billing_history::ActiveModel {
uid: Set(Uuid::new_v4()),
project: Set(project_uid),
user: Set(Some(creator_uid)),
amount: Set(first_project_credit()),
currency: Set("USD".to_string()),
reason: Set("first_project_credit".to_string()),
extra: Set(Some(serde_json::json!({
"is_first_project": true,
}))),
created_at: Set(now),
..Default::default()
}
.insert(db)
.await
.map_err(|e| AgentError::Internal(format!("failed to record credit history: {}", e)))?;
}
tracing::info!(
project_id = %project_uid,
creator_id = %creator_uid,
is_first_project = is_first,
balance = if is_first { "$20" } else { "$0" },
"project_billing_initialized"
);
Ok(())
}
// ── Internal helpers ──
async fn compute_cost(
db: &AppDatabase,
model_id: Uuid,
input_tokens: i64,
output_tokens: i64,
) -> Result<Decimal, AgentError> {
let pricing = model_pricing::Entity::find()
.filter(model_pricing::Column::ModelVersionId.eq(model_id))
.order_by_desc(model_pricing::Column::EffectiveFrom)
.one(db)
.await?
.ok_or_else(|| {
AgentError::Internal(
"No pricing record found for this model. Please configure AI model pricing first."
.into(),
)
})?;
let input_price: Decimal = pricing
.input_price_per_1k_tokens
.parse()
.map_err(|e| AgentError::Internal(format!("Invalid input price: {}", e)))?;
let output_price: Decimal = pricing
.output_price_per_1k_tokens
.parse()
.map_err(|e| AgentError::Internal(format!("Invalid output price: {}", e)))?;
if input_price <= Decimal::ZERO && output_price <= Decimal::ZERO {
return Err(AgentError::Internal(
"Model pricing is not configured or is zero. Please configure non-zero AI model pricing first."
.into(),
));
}
// DB stores per-1M-token prices; divide tokens by 1M to compute cost.
let million = Decimal::from(1_000_000);
Ok((Decimal::from(input_tokens) / million) * input_price
+ (Decimal::from(output_tokens) / million) * output_price)
}
async fn get_currency(db: &AppDatabase, model_id: Uuid) -> Result<String, AgentError> {
let pricing = model_pricing::Entity::find()
.filter(model_pricing::Column::ModelVersionId.eq(model_id))
.one(db)
.await?
.ok_or_else(|| AgentError::Internal("No pricing found".into()))?;
Ok(pricing.currency.clone())
}
async fn get_project_balance(db: &AppDatabase, project_uid: Uuid) -> Decimal {
project_billing::Entity::find_by_id(project_uid)
.one(db)
.await
.ok()
.flatten()
.map(|b| b.balance)
.unwrap_or(Decimal::ZERO)
}
async fn get_user_balance(db: &AppDatabase, user_uid: Uuid) -> Decimal {
user_billing::Entity::find_by_id(user_uid)
.one(db)
.await
.ok()
.flatten()
.map(|b| b.balance)
.unwrap_or(Decimal::ZERO)
}
async fn deduct_from_project(
db: &AppDatabase,
project_uid: Uuid,
cost: Decimal,
currency: &str,
model_id: Uuid,
input_tokens: i64,
output_tokens: i64,
) -> Result<(), String> {
let txn = db
.begin()
.await
.map_err(|e| format!("db txn error: {}", e))?;
let billing = project_billing::Entity::find_by_id(project_uid)
.lock_exclusive()
.one(&txn)
.await
.map_err(|e| format!("db error: {}", e))?
.ok_or_else(|| "Project billing account not found".to_string())?;
if billing.balance < cost {
txn.rollback().await.ok();
return Err(format!(
"Project balance insufficient. Required: {:.4} {}, Available: {:.4} {}",
cost, currency, billing.balance, currency
));
}
let now = chrono::Utc::now();
project_billing_history::ActiveModel {
uid: Set(Uuid::new_v4()),
project: Set(project_uid),
user: Set(None),
amount: Set(-cost),
currency: Set(currency.to_string()),
reason: Set("ai_usage".to_string()),
extra: Set(Some(serde_json::json!({
"model_id": model_id.to_string(),
"input_tokens": input_tokens,
"output_tokens": output_tokens,
"deducted_from": "project",
}))),
created_at: Set(now),
..Default::default()
}
.insert(&txn)
.await
.map_err(|e| format!("failed to insert history: {}", e))?;
let mut updated: project_billing::ActiveModel = billing.into();
updated.balance = Set(updated.balance.unwrap() - cost);
updated.updated_at = Set(now);
updated
.update(&txn)
.await
.map_err(|e| format!("failed to update balance: {}", e))?;
txn.commit()
.await
.map_err(|e| format!("commit error: {}", e))?;
Ok(())
}
async fn deduct_from_user(
db: &AppDatabase,
user_uid: Uuid,
cost: Decimal,
currency: &str,
project_uid: Uuid,
model_id: Uuid,
input_tokens: i64,
output_tokens: i64,
) -> Result<(), String> {
let txn = db
.begin()
.await
.map_err(|e| format!("db txn error: {}", e))?;
let billing = user_billing::Entity::find_by_id(user_uid)
.lock_exclusive()
.one(&txn)
.await
.map_err(|e| format!("db error: {}", e))?
.ok_or_else(|| "User billing account not found".to_string())?;
if billing.balance < cost {
txn.rollback().await.ok();
return Err(format!(
"Insufficient balance (project + user). Project: unavailable, User: {:.4} {}. Required: {:.4} {}",
billing.balance, currency, cost, currency
));
}
let now = chrono::Utc::now();
// Record in project billing history (but deducted from user)
project_billing_history::ActiveModel {
uid: Set(Uuid::new_v4()),
project: Set(project_uid),
user: Set(Some(user_uid)),
amount: Set(-cost),
currency: Set(currency.to_string()),
reason: Set("ai_usage_user_fallback".to_string()),
extra: Set(Some(serde_json::json!({
"model_id": model_id.to_string(),
"input_tokens": input_tokens,
"output_tokens": output_tokens,
"deducted_from": "user",
}))),
created_at: Set(now),
..Default::default()
}
.insert(&txn)
.await
.map_err(|e| format!("failed to insert history: {}", e))?;
let mut updated: user_billing::ActiveModel = billing.into();
updated.balance = Set(updated.balance.unwrap() - cost);
updated.updated_at = Set(now);
updated
.update(&txn)
.await
.map_err(|e| format!("failed to update user balance: {}", e))?;
txn.commit()
.await
.map_err(|e| format!("commit error: {}", e))?;
Ok(())
}
async fn deduct_from_user_personal(
db: &AppDatabase,
user_uid: Uuid,
cost: Decimal,
currency: &str,
model_id: Uuid,
input_tokens: i64,
output_tokens: i64,
) -> Result<(), String> {
let txn = db
.begin()
.await
.map_err(|e| format!("db txn error: {}", e))?;
let billing = user_billing::Entity::find_by_id(user_uid)
.lock_exclusive()
.one(&txn)
.await
.map_err(|e| format!("db error: {}", e))?
.ok_or_else(|| "User billing account not found".to_string())?;
if billing.balance < cost {
txn.rollback().await.ok();
return Err(format!(
"Insufficient balance. User: {:.4} {}. Required: {:.4} {}",
billing.balance, billing.currency, cost, billing.currency
));
}
let now = chrono::Utc::now();
user_billing_history::ActiveModel {
uid: Set(Uuid::new_v4()),
user: Set(user_uid),
amount: Set(-cost),
currency: Set(currency.to_string()),
reason: Set("ai_usage_personal".to_string()),
extra: Set(Some(serde_json::json!({
"model_id": model_id.to_string(),
"input_tokens": input_tokens,
"output_tokens": output_tokens,
"deducted_from": "user",
"scope": "personal",
}))),
created_at: Set(now),
..Default::default()
}
.insert(&txn)
.await
.map_err(|e| format!("failed to insert user history: {}", e))?;
let mut updated: user_billing::ActiveModel = billing.into();
updated.balance = Set(updated.balance.unwrap() - cost);
updated.updated_at = Set(now);
updated
.update(&txn)
.await
.map_err(|e| format!("failed to update user balance: {}", e))?;
txn.commit()
.await
.map_err(|e| format!("commit error: {}", e))?;
Ok(())
}
pub async fn persist_billing_error(
db: &AppDatabase,
scope: &str,
scope_id: Uuid,
error_type: &str,
message: &str,
details: Option<serde_json::Value>,
) -> Result<(), AgentError> {
billing_error::ActiveModel {
id: Set(Uuid::new_v4()),
scope: Set(scope.to_string()),
scope_id: Set(scope_id),
error_type: Set(error_type.to_string()),
message: Set(message.to_string()),
details: Set(details),
resolved: Set(false),
created_at: Set(chrono::Utc::now()),
}
.insert(db)
.await
.map_err(|e| AgentError::Internal(format!("failed to persist billing error: {}", e)))?;
tracing::warn!(scope, %scope_id, error_type, "billing_error_persisted");
Ok(())
}
fn decimal_to_f64(d: Decimal) -> f64 {
d.round_dp(10).to_string().parse().unwrap_or(0.0)
}

1256
libs/agent/chat/chat_execution.rs
View File

File diff suppressed because it is too large Load Diff

163
libs/agent/chat/mod.rs
View File

@ -1,163 +0,0 @@
use std::pin::Pin;
use config::AppConfig;
use db::cache::AppCache;
use db::database::AppDatabase;
use models::agents::model;
use models::projects::{project, project_context_setting};
use models::repos::repo;
use models::rooms::{room, room_message};
use models::users::user;
use std::collections::HashMap;
use uuid::Uuid;
/// Maximum recursion rounds for tool-call loops (AI → tool → result → AI).
/// Previous default of 3 caused frequent silent termination on realistic multi-step queries.
pub const DEFAULT_MAX_TOOL_DEPTH: usize = 99;
/// A single chunk from an AI streaming response.
#[derive(Debug, Clone)]
pub struct AiStreamChunk {
pub content: String,
pub done: bool,
/// What kind of content this chunk contains — helps the frontend render
/// thinking, tool calls, and results with different styles.
pub chunk_type: AiChunkType,
/// Structured metadata for tool_call / tool_result events.
/// tool_call: {"tool": "...", "args": {...}}
/// tool_result: {"tool": "...", "status": "ok|error", "result": "..."}
pub metadata: Option<serde_json::Value>,
/// Optional ID of a child process/agent, sent to frontend via SSE.
pub children_id: Option<String>,
}
/// Type of streaming chunk, used by the frontend for rendering.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum AiChunkType {
/// AI reasoning/thinking text before a tool call or answer.
Thinking,
/// Final answer text from the AI.
Answer,
/// A tool call is being executed (content = tool name + args summary).
ToolCall,
/// Tool execution result (content = result or error).
ToolResult,
}
impl Default for AiChunkType {
fn default() -> Self {
Self::Answer
}
}
const THINK_OPEN: &str = "\x3cthinking\x3e";
const THINK_CLOSE: &str = "\x3c/response\x3e";
/// Strip XML-format thinking tags that some models (e.g. DeepSeek-R1) embed
/// in reasoning output. Also normalizes excessive consecutive newlines (3+ → 2).
pub fn normalize_thinking_content(content: &str) -> String {
let content = content
.replace(THINK_CLOSE, "")
.replace(THINK_OPEN, "")
.replace("\x3cthinking", "")
.replace("/response\x3e", "");
let mut result = String::with_capacity(content.len());
let mut newline_count = 0usize;
for ch in content.chars() {
if ch == '\n' {
newline_count += 1;
if newline_count <= 2 {
result.push(ch);
}
} else {
newline_count = 0;
result.push(ch);
}
}
result.trim().to_string()
}
pub type StreamCallback = Box<
dyn Fn(AiStreamChunk) -> Pin<Box<dyn std::future::Future<Output = ()> + Send>> + Send + Sync,
>;
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum AgentRole {
Default,
Supervisor,
Researcher,
Analyst,
Reviewer,
Architect,
Debugger,
Implementer,
Tester,
Security,
}
#[derive(Debug, Clone, Default)]
pub struct AgentExecutionProfile {
pub role: AgentRole,
pub system_prompt: Option<String>,
pub temperature: Option<f64>,
pub max_tokens: Option<i32>,
pub top_p: Option<f64>,
pub frequency_penalty: Option<f64>,
pub presence_penalty: Option<f64>,
pub max_tool_depth: Option<usize>,
pub allowed_tools: Option<Vec<String>>,
pub disable_orchestration: bool,
}
impl Default for AgentRole {
fn default() -> Self {
Self::Default
}
}
#[derive(Clone)]
pub struct AiRequest {
pub db: AppDatabase,
pub cache: AppCache,
pub config: AppConfig,
pub model: model::Model,
pub project: project::Model,
pub context_setting: Option<project_context_setting::Model>,
pub sender: user::Model,
pub room: room::Model,
pub input: String,
pub mention: Vec<Mention>,
pub history: Vec<room_message::Model>,
pub history_cutoff_seq: Option<i64>,
pub user_names: HashMap<Uuid, String>,
pub temperature: f64,
pub max_tokens: i32,
pub top_p: f64,
pub frequency_penalty: f64,
pub presence_penalty: f64,
pub think: bool,
pub tools: Option<Vec<serde_json::Value>>,
pub max_tool_depth: usize,
pub execution_profile: Option<AgentExecutionProfile>,
pub room_preamble: Option<String>,
}
#[derive(Clone)]
pub enum Mention {
User(user::Model),
Repo(repo::Model),
}
pub mod agent_profile;
pub mod chat_execution;
pub mod context;
pub mod message_builder;
pub mod nonstreaming_execution;
pub mod orchestrator;
pub mod react_execution;
pub mod service;
pub mod session_recording;
pub mod state;
pub mod streaming_execution;
pub use context::{AiContextSenderType, RoomMessageContext};
pub use service::ChatService;
pub use state::{AgentRuntime, AgentState};

317
libs/agent/chat/orchestrator.rs
View File

@ -1,317 +0,0 @@
use std::collections::HashMap;
use super::agent_profile::{profile_for_role_name, should_enable_delegation, supervisor_profile};
use super::message_builder::MessageBuilder;
use super::nonstreaming_execution::execute_process;
use super::service::{ProcessResult, StreamResult};
use super::{AiRequest, StreamCallback};
use crate::error::Result;
use crate::tool::call::ToolError;
use crate::tool::registry::ToolRegistry;
use crate::tool::{ToolDefinition, ToolHandler, ToolParam, ToolSchema};
pub async fn execute_orchestrated_process(
request: AiRequest,
message_builder: &MessageBuilder,
tool_registry: &Option<ToolRegistry>,
ai_base_url: Option<String>,
ai_api_key: Option<String>,
) -> Result<ProcessResult> {
if request
.execution_profile
.as_ref()
.is_some_and(|p| p.disable_orchestration)
{
return execute_process(
request,
message_builder,
tool_registry,
ai_base_url,
ai_api_key,
)
.await;
}
let tools = request.tools.clone().unwrap_or_default();
if !should_enable_delegation(&request.input, !tools.is_empty()) {
return execute_process(
request,
message_builder,
tool_registry,
ai_base_url,
ai_api_key,
)
.await;
}
let mut enhanced_registry = tool_registry.clone().unwrap_or_default();
register_call_sub_agent_tool(
&mut enhanced_registry,
&request,
message_builder,
tool_registry,
ai_base_url.clone(),
ai_api_key.clone(),
);
let mut supervisor_request = request.clone();
let profile = supervisor_profile();
supervisor_request.execution_profile = Some(profile.clone());
supervisor_request.tools = Some(enhanced_registry.to_openai_tools());
supervisor_request.temperature = profile.temperature.unwrap_or(request.temperature);
supervisor_request.max_tokens = profile.max_tokens.unwrap_or(request.max_tokens);
supervisor_request.top_p = profile.top_p.unwrap_or(request.top_p);
supervisor_request.frequency_penalty = profile
.frequency_penalty
.unwrap_or(request.frequency_penalty);
supervisor_request.presence_penalty =
profile.presence_penalty.unwrap_or(request.presence_penalty);
execute_process(
supervisor_request,
message_builder,
&Some(enhanced_registry),
ai_base_url,
ai_api_key,
)
.await
}
pub async fn execute_orchestrated_stream(
request: AiRequest,
on_chunk: StreamCallback,
message_builder: &MessageBuilder,
tool_registry: &Option<ToolRegistry>,
ai_base_url: Option<String>,
ai_api_key: Option<String>,
) -> Result<StreamResult> {
if request
.execution_profile
.as_ref()
.is_some_and(|p| p.disable_orchestration)
{
return super::streaming_execution::execute_process_stream(
request,
on_chunk,
message_builder,
tool_registry,
ai_base_url,
ai_api_key,
)
.await;
}
let tools = request.tools.clone().unwrap_or_default();
if !should_enable_delegation(&request.input, !tools.is_empty()) {
return super::streaming_execution::execute_process_stream(
request,
on_chunk,
message_builder,
tool_registry,
ai_base_url,
ai_api_key,
)
.await;
}
let mut enhanced_registry = tool_registry.clone().unwrap_or_default();
register_call_sub_agent_tool(
&mut enhanced_registry,
&request,
message_builder,
tool_registry,
ai_base_url.clone(),
ai_api_key.clone(),
);
let mut supervisor_request = request.clone();
let profile = supervisor_profile();
supervisor_request.execution_profile = Some(profile.clone());
supervisor_request.tools = Some(enhanced_registry.to_openai_tools());
supervisor_request.temperature = profile.temperature.unwrap_or(request.temperature);
supervisor_request.max_tokens = profile.max_tokens.unwrap_or(request.max_tokens);
supervisor_request.top_p = profile.top_p.unwrap_or(request.top_p);
supervisor_request.frequency_penalty = profile
.frequency_penalty
.unwrap_or(request.frequency_penalty);
supervisor_request.presence_penalty =
profile.presence_penalty.unwrap_or(request.presence_penalty);
super::streaming_execution::execute_process_stream(
supervisor_request,
on_chunk,
message_builder,
&Some(enhanced_registry),
ai_base_url,
ai_api_key,
)
.await
}
fn register_call_sub_agent_tool(
registry: &mut ToolRegistry,
request: &AiRequest,
message_builder: &MessageBuilder,
original_registry: &Option<ToolRegistry>,
ai_base_url: Option<String>,
ai_api_key: Option<String>,
) {
let captured_request = request.clone();
let captured_message_builder = message_builder.clone();
let captured_original_registry = original_registry.clone();
let captured_base_url = ai_base_url;
let captured_api_key = ai_api_key;
registry.register(
ToolDefinition::new("call_sub_agent")
.description(
"Delegate a task to a specialist sub-agent and receive its output.\n\
Available roles:\n\
- researcher: Gathers facts, evidence, and data. Best for finding information and searching code.\n\
- analyst: Builds explanations, highlights causal links and tradeoffs. Best for reasoning about implications.\n\
- reviewer: Stress-tests proposals, identifies risks and contradictions. Best for quality checks.\n\
- architect: Maps systems, dependencies, boundaries, and design tradeoffs. Best for architecture decisions.\n\
- debugger: Finds root causes, suspect changes, and validation paths. Best for bugs and regressions.\n\
- implementer: Converts requirements into concrete implementation steps. Best for execution planning.\n\
- tester: Designs validation and regression coverage. Best for test strategy.\n\
- security: Reviews auth, data exposure, injection, dependency, and abuse risks. Best for sensitive changes.\n\
Provide a clear, focused task description so the sub-agent knows exactly what to investigate.",
)
.parameters(ToolSchema {
schema_type: "object".into(),
properties: Some({
let mut p = HashMap::new();
p.insert(
"role".into(),
ToolParam {
name: "role".into(),
param_type: "string".into(),
description: Some(
"The sub-agent role to delegate to: researcher, analyst, reviewer, architect, debugger, implementer, tester, or security.".into(),
),
required: true,
properties: None,
items: None,
},
);
p.insert(
"task".into(),
ToolParam {
name: "task".into(),
param_type: "string".into(),
description: Some(
"The specific task or question for the sub-agent. Be precise and focused.".into(),
),
required: true,
properties: None,
items: None,
},
);
p
}),
required: Some(vec!["role".into(), "task".into()]),
}),
ToolHandler::new(move |_ctx, args| {
// Extract owned values from args before async move (avoid borrowing across boundary)
let role = args
.get("role")
.and_then(|v| v.as_str())
.unwrap_or("researcher")
.to_owned();
let task = args
.get("task")
.and_then(|v| v.as_str())
.unwrap_or("")
.to_owned();
let profile = profile_for_role_name(role.as_str());
let mut sub_request = captured_request.clone();
sub_request.input = format!(
"Sub-agent role: {role}\n\nTask:\n{task}\n\nOriginal user request:\n{}\n\nInstructions:\nFocus only on your assigned task. Return concise, evidence-backed findings.",
captured_request.input
);
sub_request.execution_profile = Some(profile.clone());
sub_request.tools = Some(filter_tools_for_sub_agent(
&captured_request.tools,
&profile.allowed_tools,
));
sub_request.max_tool_depth = profile
.max_tool_depth
.unwrap_or(captured_request.max_tool_depth);
sub_request.temperature = profile.temperature.unwrap_or(captured_request.temperature);
sub_request.max_tokens = profile.max_tokens.unwrap_or(captured_request.max_tokens);
sub_request.top_p = profile.top_p.unwrap_or(captured_request.top_p);
sub_request.frequency_penalty = profile
.frequency_penalty
.unwrap_or(captured_request.frequency_penalty);
sub_request.presence_penalty = profile
.presence_penalty
.unwrap_or(captured_request.presence_penalty);
// Clone captured values for this invocation so the Fn closure retains them
let mb = captured_message_builder.clone();
let sub_registry = captured_original_registry.clone();
let base = captured_base_url.clone();
let key = captured_api_key.clone();
Box::pin(async move {
let result = execute_process(sub_request, &mb, &sub_registry, base, key).await;
match result {
Ok(r) => Ok(serde_json::json!({
"role": role,
"output": r.content,
"input_tokens": r.input_tokens,
"output_tokens": r.output_tokens,
})),
Err(e) => Err(ToolError::ExecutionError(format!(
"Sub-agent '{}' execution failed: {}",
role, e
))),
}
})
}),
);
}
/// Filter the original tool definitions by the sub-agent's allowed list,
/// always excluding `call_sub_agent` to prevent recursive delegation.
fn filter_tools_for_sub_agent(
original_tools: &Option<Vec<serde_json::Value>>,
allowed_tools: &Option<Vec<String>>,
) -> Vec<serde_json::Value> {
let Some(tools) = original_tools else {
return Vec::new();
};
let allowed = allowed_tools.as_ref().map(|list| {
list.iter()
.filter(|n| *n != "call_sub_agent")
.cloned()
.collect::<Vec<String>>()
});
match allowed {
Some(allowed_list) if !allowed_list.is_empty() => tools
.iter()
.filter(|tool| {
let name = tool
.get("function")
.and_then(|f| f.get("name"))
.and_then(|v| v.as_str())
.unwrap_or("");
allowed_list.iter().any(|allowed| allowed == name)
})
.cloned()
.collect(),
_ => tools
.iter()
.filter(|tool| {
tool.get("function")
.and_then(|f| f.get("name"))
.and_then(|v| v.as_str())
.is_some_and(|name| name != "call_sub_agent")
})
.cloned()
.collect(),
}
}

233
libs/agent/chat/react_execution.rs
View File

@ -1,233 +0,0 @@
use futures::StreamExt;
use models::rooms::room_ai;
use rig::agent::{AgentBuilder, MultiTurnStreamItem};
use rig::client::CompletionClient;
use rig::streaming::{StreamedAssistantContent, StreamingPrompt};
use sea_orm::*;
use uuid::Uuid;
use super::AiRequest;
use super::session_recording::record_ai_session;
use crate::client::AiClientConfig;
use crate::error::{AgentError, Result};
use crate::react::types::Action as ReactAction;
use crate::react::{DEFAULT_SYSTEM_PROMPT, ReactStep};
use crate::tool::{RecordingTool, registry::ToolRegistry};
pub async fn execute_process_react<C, Fut>(
request: &AiRequest,
mut on_chunk: C,
tool_registry: &ToolRegistry,
ai_base_url: Option<String>,
ai_api_key: Option<String>,
room_preamble: Option<&str>,
message_producer: Option<queue::MessageProducer>,
) -> Result<(String, i64, i64)>
where
C: FnMut(ReactStep) -> Fut + Send,
Fut: std::future::Future<Output = ()> + Send,
{
let base_url = ai_base_url.unwrap_or_else(|| "https://api.openai.com".into());
let api_key = ai_api_key.unwrap_or_default();
let client_config = AiClientConfig::new(api_key).with_base_url(base_url);
let db = request.db.clone();
let cache = request.cache.clone();
let cfg = request.config.clone();
let room_id = request.room.id;
let sender_uid = request.sender.uid;
let project_id = request.project.id;
let ai_model_id = request.model.id;
let ai_model_name = request.model.name.clone();
let sent_in_turn = std::sync::Arc::new(std::sync::Mutex::new(Vec::new()));
let session_id = Uuid::now_v7();
let session_start = std::time::Instant::now();
let version_id = room_ai::Entity::find()
.filter(room_ai::Column::Room.eq(request.room.id))
.filter(room_ai::Column::Model.eq(request.model.id))
.one(&request.db)
.await
.ok()
.flatten()
.and_then(|r| r.version);
let mut tools: Vec<Box<dyn rig::tool::ToolDyn + 'static>> = Vec::new();
for def in tool_registry.definitions() {
let name = def.name.clone();
if let Some(handler) = tool_registry.get(&name) {
let adapter = crate::tool::RigToolAdapter::new(
handler.clone(),
def.clone(),
db.clone(),
cache.clone(),
cfg.clone(),
room_id,
Some(sender_uid),
project_id,
message_producer.clone(),
Some(ai_model_id),
Some(ai_model_name.clone()),
sent_in_turn.clone(),
);
tools.push(Box::new(RecordingTool::new(
Box::new(adapter),
db.clone(),
session_id,
sender_uid,
)));
}
}
let rig_client = client_config.build_rig_client();
let model = rig_client.completion_model(&request.model.name);
// General rules first (strong LLM attention), room context appended after
// so that output-format rules aren't buried behind long room preamble.
let preamble = match room_preamble {
Some(rp) => format!("{}\n{}", DEFAULT_SYSTEM_PROMPT, rp),
None => DEFAULT_SYSTEM_PROMPT.to_string(),
};
let agent = AgentBuilder::new(model)
.preamble(&preamble)
.tools(tools)
.default_max_turns(request.max_tool_depth)
.build();
let stream = agent
.stream_prompt(&request.input)
.with_history(Vec::<rig::completion::Message>::new())
.multi_turn(request.max_tool_depth)
.await;
tokio::pin!(stream);
let mut step_count = 0usize;
let mut final_content = String::new();
let mut total_input_tokens: i64 = 0;
let mut total_output_tokens: i64 = 0;
while let Some(item) = stream.next().await {
match item {
Ok(MultiTurnStreamItem::StreamAssistantItem(StreamedAssistantContent::Text(text))) => {
step_count += 1;
let t = text.text;
on_chunk(ReactStep::Answer {
step: step_count,
answer: t.clone(),
})
.await;
final_content.push_str(&t);
}
Ok(MultiTurnStreamItem::StreamAssistantItem(StreamedAssistantContent::Reasoning(
reasoning,
))) => {
let reasoning_text: String = reasoning
.content
.iter()
.filter_map(|c| match c {
rig::completion::message::ReasoningContent::Text { text, .. } => {
Some(text.as_str())
}
_ => None,
})
.collect::<Vec<_>>()
.join("");
if !reasoning_text.is_empty() {
step_count += 1;
on_chunk(ReactStep::Thought {
step: step_count,
thought: reasoning_text,
})
.await;
}
}
Ok(MultiTurnStreamItem::StreamAssistantItem(
StreamedAssistantContent::ReasoningDelta { reasoning, .. },
)) => {
if !reasoning.is_empty() {
step_count += 1;
on_chunk(ReactStep::Thought {
step: step_count,
thought: reasoning,
})
.await;
}
}
Ok(MultiTurnStreamItem::StreamAssistantItem(StreamedAssistantContent::ToolCall {
tool_call,
..
})) => {
step_count += 1;
let args: serde_json::Value = match &tool_call.function.arguments {
serde_json::Value::String(s) => {
serde_json::from_str(s).unwrap_or(serde_json::Value::Null)
}
v => v.clone(),
};
on_chunk(ReactStep::Action {
step: step_count,
action: ReactAction::new(&tool_call.function.name, args),
})
.await;
}
Ok(MultiTurnStreamItem::StreamUserItem(
rig::streaming::StreamedUserContent::ToolResult { tool_result, .. },
)) => {
step_count += 1;
let obs = tool_result_content_to_string(&tool_result.content);
on_chunk(ReactStep::Observation {
step: step_count,
observation: obs,
})
.await;
}
Ok(MultiTurnStreamItem::FinalResponse(resp)) => {
let usage = resp.usage();
total_input_tokens = usage.input_tokens as i64;
total_output_tokens = usage.output_tokens as i64;
}
Err(e) => {
let err_msg = format!("rig agent stream error: {}", e);
return Err(AgentError::OpenAi(err_msg));
}
_ => {}
}
}
let elapsed_ms = session_start.elapsed().as_millis() as i64;
record_ai_session(
&request.cache,
&request.db,
request.project.id,
request.sender.uid,
session_id,
request.room.id,
request.model.id,
version_id.unwrap_or_default(),
total_input_tokens,
total_output_tokens,
elapsed_ms,
)
.await;
Ok((final_content, total_input_tokens, total_output_tokens))
}
/// Extract text from rig's ToolResultContent, ignoring images.
fn tool_result_content_to_string(
content: &rig::one_or_many::OneOrMany<rig::completion::message::ToolResultContent>,
) -> String {
use rig::completion::message::ToolResultContent;
content
.iter()
.filter_map(|item| {
if let ToolResultContent::Text(t) = item {
Some(t.text.clone())
} else {
None
}
})
.collect::<Vec<_>>()
.join("\n")
}

268
libs/agent/chat/service.rs
View File

@ -1,268 +0,0 @@
use super::message_builder::MessageBuilder;
use super::{AiRequest, StreamCallback};
use crate::client::AiClientConfig;
use crate::client::StreamChunk;
use crate::compact::CompactService;
use crate::embed::EmbedService;
use crate::error::Result;
use crate::perception::PerceptionService;
use crate::tool::registry::ToolRegistry;
use queue::MessageProducer;
/// Result from streaming AI response.
pub struct StreamResult {
pub content: String,
pub reasoning_content: String,
pub input_tokens: i64,
pub output_tokens: i64,
/// All chunks in arrival order — preserves ReAct multi-cycle ordering.
pub chunks: Vec<StreamChunk>,
}
/// Result from non-streaming AI response.
pub struct ProcessResult {
pub content: String,
pub input_tokens: i64,
pub output_tokens: i64,
}
/// Service for handling AI chat requests in rooms.
pub struct ChatService {
ai_base_url: Option<String>,
ai_api_key: Option<String>,
message_builder: MessageBuilder,
tool_registry: Option<ToolRegistry>,
}
impl ChatService {
pub fn new() -> Self {
Self {
ai_base_url: None,
ai_api_key: None,
message_builder: MessageBuilder::new(),
tool_registry: None,
}
}
pub fn with_ai_client_config(mut self, config: AiClientConfig) -> Self {
self.ai_base_url = config.base_url.clone();
self.ai_api_key = Some(config.api_key.clone());
self
}
pub fn with_compact_service(mut self, compact_service: CompactService) -> Self {
self.message_builder = self.message_builder.with_compact_service(compact_service);
self
}
pub fn with_embed_service(mut self, embed_service: EmbedService) -> Self {
self.message_builder = self.message_builder.with_embed_service(embed_service);
self
}
pub fn with_perception_service(mut self, perception_service: PerceptionService) -> Self {
self.message_builder = self
.message_builder
.with_perception_service(perception_service);
self
}
pub fn with_tool_registry(mut self, registry: ToolRegistry) -> Self {
self.tool_registry = Some(registry);
self
}
/// Returns all registered tools as JSON tool definitions.
pub fn tools(&self) -> Vec<serde_json::Value> {
self.tool_registry
.as_ref()
.map(|r| r.to_openai_tools())
.unwrap_or_default()
}
/// Build a RigToolSet from the registered tool registry.
///
/// This enables using the same tools with `RigAgentService` via rig's native Agent.
/// The context (db, cache, config, room_id, sender_id) is passed through to each
/// tool handler at creation time.
#[cfg(feature = "rig")]
pub fn rig_toolset(
&self,
db: db::database::AppDatabase,
cache: db::cache::AppCache,
config: config::AppConfig,
room_id: uuid::Uuid,
sender_id: Option<uuid::Uuid>,
project_id: uuid::Uuid,
) -> Option<crate::RigToolSet> {
self.tool_registry.as_ref().map(|registry| {
crate::RigToolSet::from_registry(
registry,
db,
cache,
config,
room_id,
sender_id,
project_id,
None,
None,
None,
std::sync::Arc::new(std::sync::Mutex::new(Vec::new())),
)
})
}
/// Get a reference to the underlying ToolRegistry.
pub fn tool_registry(&self) -> Option<&ToolRegistry> {
self.tool_registry.as_ref()
}
pub async fn build_room_optimized_context_text(
&self,
request: &AiRequest,
) -> Result<(String, Option<i64>)> {
self.message_builder
.build_room_optimized_context_text(request)
.await
}
/// Process AI request without streaming (tool-call loop with non-streaming API).
pub async fn process(&self, request: AiRequest) -> Result<ProcessResult> {
super::orchestrator::execute_orchestrated_process(
request,
&self.message_builder,
&self.tool_registry,
self.ai_base_url.clone(),
self.ai_api_key.clone(),
)
.await
}
/// Process AI request with streaming (tool-call loop with streaming API, incremental chunks).
pub async fn process_stream(
&self,
request: AiRequest,
on_chunk: StreamCallback,
) -> Result<StreamResult> {
super::orchestrator::execute_orchestrated_stream(
request,
on_chunk,
&self.message_builder,
&self.tool_registry,
self.ai_base_url.clone(),
self.ai_api_key.clone(),
)
.await
}
/// Process AI request for room context — direct execution path (bypasses orchestrator).
///
/// Room AI uses a fast single-agent loop: all tools available, no multi-agent delegation.
/// Merges `room_tools` (send_message, retract_message) into the base registry,
/// then runs `execute_process` / `execute_process_stream` directly.
pub async fn process_room(
&self,
request: AiRequest,
room_tools: ToolRegistry,
) -> Result<ProcessResult> {
let mut merged = self.tool_registry.clone().unwrap_or_default();
merged.merge(room_tools);
super::nonstreaming_execution::execute_process(
request,
&self.message_builder,
&Some(merged),
self.ai_base_url.clone(),
self.ai_api_key.clone(),
)
.await
}
/// Process AI request for room context with streaming — direct execution path.
///
/// Same as `process_room` but with streaming response. Bypasses orchestrator,
/// gives the room AI all tools (base + room) for fast single-agent execution.
pub async fn process_room_stream(
&self,
request: AiRequest,
on_chunk: StreamCallback,
room_tools: ToolRegistry,
) -> Result<StreamResult> {
let mut merged = self.tool_registry.clone().unwrap_or_default();
merged.merge(room_tools);
super::streaming_execution::execute_process_stream(
request,
on_chunk,
&self.message_builder,
&Some(merged),
self.ai_base_url.clone(),
self.ai_api_key.clone(),
)
.await
}
/// Process AI request via rig-based ReAct streaming loop.
pub async fn process_react<C, Fut>(
&self,
request: &AiRequest,
on_chunk: C,
) -> Result<(String, i64, i64)>
where
C: FnMut(crate::react::ReactStep) -> Fut + Send,
Fut: std::future::Future<Output = ()> + Send,
{
let Some(registry) = &self.tool_registry else {
return Err(crate::error::AgentError::Internal(
"no tool registry registered".into(),
));
};
super::react_execution::execute_process_react(
request,
on_chunk,
registry,
self.ai_base_url.clone(),
self.ai_api_key.clone(),
None,
None,
)
.await
}
/// Process AI request via rig-based ReAct streaming loop with room-specific tools.
///
/// Merges `room_tools` (e.g. `send_message`, `retract_message`) into the base
/// tool registry on-the-fly. The `room_preamble` is prepended to the default
/// system prompt to instruct the AI about room communication rules.
/// `message_producer` enables tools to publish events via the message queue.
pub async fn process_react_room<C, Fut>(
&self,
request: &AiRequest,
on_chunk: C,
room_tools: ToolRegistry,
room_preamble: Option<&str>,
message_producer: Option<MessageProducer>,
) -> Result<(String, i64, i64)>
where
C: FnMut(crate::react::ReactStep) -> Fut + Send,
Fut: std::future::Future<Output = ()> + Send,
{
let Some(registry) = &self.tool_registry else {
return Err(crate::error::AgentError::Internal(
"no tool registry registered".into(),
));
};
let mut merged = registry.clone();
merged.merge(room_tools);
super::react_execution::execute_process_react(
request,
on_chunk,
&merged,
self.ai_base_url.clone(),
self.ai_api_key.clone(),
room_preamble,
message_producer,
)
.await
}
}

217
libs/agent/chat/state.rs
View File

@ -1,217 +0,0 @@
//! Agent state machine — tracks lifecycle of a single AI agent invocation.
//!
//! States: Idle → Thinking → ToolCall → Thinking → ... → Answering | Error
//! The Thinking ↔ ToolCall cycle repeats until max tool depth or final answer.
use chrono::{DateTime, Utc};
use serde::{Deserialize, Serialize};
/// Current phase of an agent's execution lifecycle.
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub enum AgentState {
/// Agent is idle, waiting for input
Idle,
/// Agent is reasoning/thinking (may produce thinking chunks)
Thinking {
started_at: DateTime<Utc>,
tool_depth: u32,
},
/// Agent is executing a tool call
ToolCall {
tool_name: String,
started_at: DateTime<Utc>,
},
/// Agent is returning the final answer
Answering {
/// Accumulated answer content so far
content_chars: u64,
started_at: DateTime<Utc>,
},
/// Agent encountered a non-recoverable error
Error { message: String, tool_depth: u32 },
}
impl AgentState {
pub fn is_terminal(&self) -> bool {
matches!(
self,
AgentState::Answering { .. } | AgentState::Error { .. }
)
}
pub fn is_idle(&self) -> bool {
matches!(self, AgentState::Idle)
}
pub fn current_phase(&self) -> &'static str {
match self {
AgentState::Idle => "idle",
AgentState::Thinking { .. } => "thinking",
AgentState::ToolCall { .. } => "tool_call",
AgentState::Answering { .. } => "answering",
AgentState::Error { .. } => "error",
}
}
}
/// State machine for agent lifecycle transitions.
pub struct AgentRuntime {
state: AgentState,
max_tool_depth: u32,
current_depth: u32,
}
impl AgentRuntime {
pub fn new(max_tool_depth: u32) -> Self {
Self {
state: AgentState::Idle,
max_tool_depth,
current_depth: 0,
}
}
pub fn state(&self) -> &AgentState {
&self.state
}
/// Transition from Idle → Thinking
pub fn start_thinking(&mut self) {
debug_assert!(self.state.is_idle(), "must be Idle to start thinking");
self.current_depth = 0;
self.state = AgentState::Thinking {
started_at: Utc::now(),
tool_depth: 0,
};
}
/// Transition from Thinking → ToolCall (increments tool depth)
pub fn start_tool_call(&mut self, tool_name: String) -> Result<(), &'static str> {
if !matches!(self.state, AgentState::Thinking { .. }) {
return Err("must be Thinking to start tool call");
}
if self.current_depth >= self.max_tool_depth {
return Err("max tool depth reached");
}
self.state = AgentState::ToolCall {
tool_name,
started_at: Utc::now(),
};
Ok(())
}
/// Transition from ToolCall → Thinking (back after tool result)
pub fn complete_tool_call(&mut self) -> Result<(), &'static str> {
if !matches!(self.state, AgentState::ToolCall { .. }) {
return Err("must be ToolCall to complete");
}
self.current_depth += 1;
self.state = AgentState::Thinking {
started_at: Utc::now(),
tool_depth: self.current_depth,
};
Ok(())
}
/// Transition to Answering (terminal)
pub fn start_answer(&mut self) {
self.state = AgentState::Answering {
content_chars: 0,
started_at: Utc::now(),
};
}
pub fn append_answer(&mut self, content: &str) {
if let AgentState::Answering { content_chars, .. } = &mut self.state {
*content_chars += content.len() as u64;
}
}
/// Transition to Error (terminal)
pub fn fail(&mut self, message: String) {
self.state = AgentState::Error {
message,
tool_depth: self.current_depth,
};
}
}
#[cfg(test)]
mod tests {
use super::*;
#[test]
fn test_starts_idle() {
let rt = AgentRuntime::new(10);
assert!(rt.state().is_idle());
assert_eq!(rt.state().current_phase(), "idle");
}
#[test]
fn test_idle_to_thinking() {
let mut rt = AgentRuntime::new(10);
rt.start_thinking();
assert_eq!(rt.state().current_phase(), "thinking");
assert!(!rt.state().is_terminal());
}
#[test]
fn test_thinking_to_tool_call_and_back() {
let mut rt = AgentRuntime::new(10);
rt.start_thinking();
rt.start_tool_call("search".into()).unwrap();
assert_eq!(rt.state().current_phase(), "tool_call");
rt.complete_tool_call().unwrap();
assert_eq!(rt.state().current_phase(), "thinking");
}
#[test]
fn test_thinking_to_answer() {
let mut rt = AgentRuntime::new(10);
rt.start_thinking();
rt.start_answer();
assert_eq!(rt.state().current_phase(), "answering");
assert!(rt.state().is_terminal());
}
#[test]
fn test_append_answer_tracks_chars() {
let mut rt = AgentRuntime::new(10);
rt.start_thinking();
rt.start_answer();
rt.append_answer("hello");
if let AgentState::Answering { content_chars, .. } = rt.state() {
assert_eq!(*content_chars, 5);
} else {
panic!("expected Answering state");
}
}
#[test]
fn test_error_is_terminal() {
let mut rt = AgentRuntime::new(10);
rt.start_thinking();
rt.fail("something broke".into());
assert_eq!(rt.state().current_phase(), "error");
assert!(rt.state().is_terminal());
}
#[test]
fn test_transition_from_wrong_state() {
let mut rt = AgentRuntime::new(10);
// Can't start tool call from Idle
assert!(rt.start_tool_call("tool".into()).is_err());
// Can't complete tool call from Idle
assert!(rt.complete_tool_call().is_err());
}
#[test]
fn test_max_depth_rejected() {
let mut rt = AgentRuntime::new(2);
rt.start_thinking();
rt.start_tool_call("tool1".into()).unwrap();
rt.complete_tool_call().unwrap();
rt.start_tool_call("tool2".into()).unwrap();
rt.complete_tool_call().unwrap();
assert!(rt.start_tool_call("tool3".into()).is_err());
}
}

511
libs/agent/chat/streaming_execution.rs
View File

@ -1,511 +0,0 @@
use models::projects::project_skill;
use models::rooms::room_ai;
use sea_orm::{ColumnTrait, EntityTrait, QueryFilter};
use std::pin::Pin;
use std::sync::Arc;
use uuid::Uuid;
use super::message_builder::MessageBuilder;
use super::service::StreamResult;
use super::session_recording::record_ai_session;
use super::{AiChunkType, AiRequest, AiStreamChunk, StreamCallback};
use crate::client::AiClientConfig;
use crate::client::types::{ChatRequestMessage, ToolCall};
use crate::client::{StreamChunk, StreamChunkType, StreamedToolCall, call_stream};
use crate::error::Result;
use crate::perception::{SkillEntry, ToolCallEvent};
use crate::tool::{ToolCall as AgentToolCall, ToolContext, ToolExecutor};
type SharedCallback = Arc<
dyn Fn(AiStreamChunk) -> Pin<Box<dyn std::future::Future<Output = ()> + Send>> + Send + Sync,
>;
pub async fn execute_process_stream(
request: AiRequest,
on_chunk: StreamCallback,
message_builder: &MessageBuilder,
tool_registry: &Option<crate::tool::registry::ToolRegistry>,
ai_base_url: Option<String>,
ai_api_key: Option<String>,
) -> Result<StreamResult> {
let on_chunk: SharedCallback = Arc::from(on_chunk);
let tools: Vec<serde_json::Value> = request.tools.clone().unwrap_or_default();
let tools_enabled = !tools.is_empty();
let max_tool_depth = request.max_tool_depth;
let mut messages = message_builder.build_messages(&request).await?;
let room_ai_config = room_ai::Entity::find()
.filter(room_ai::Column::Room.eq(request.room.id))
.filter(room_ai::Column::Model.eq(request.model.id))
.one(&request.db)
.await?;
let model_name = request.model.name.clone();
let profile = request.execution_profile.as_ref();
let temperature = profile
.and_then(|p| p.temperature.map(|v| v as f32))
.or_else(|| {
room_ai_config
.as_ref()
.and_then(|r| r.temperature.map(|v| v as f32))
})
.unwrap_or(request.temperature as f32);
let max_tokens = profile
.and_then(|p| p.max_tokens.map(|v| v as u32))
.or_else(|| {
room_ai_config
.as_ref()
.and_then(|r| r.max_tokens.map(|v| v as u32))
})
.unwrap_or(request.max_tokens as u32);
let mut tool_depth = 0;
let mut total_input_tokens = 0i64;
let mut total_output_tokens = 0i64;
let session_id = Uuid::now_v7();
let session_start = std::time::Instant::now();
let version_id = room_ai_config.as_ref().and_then(|r| r.version);
let config = AiClientConfig::new(ai_api_key.unwrap_or_default())
.with_base_url(ai_base_url.unwrap_or_else(|| "https://api.openai.com".into()));
let mut full_content = String::new();
let mut all_chunks: Vec<StreamChunk> = Vec::new();
let (tx, mut rx) = tokio::sync::mpsc::unbounded_channel::<StreamedToolCall>();
loop {
let on_chunk_cb = on_chunk.clone();
let on_chunk_cb2 = on_chunk.clone();
let tx_arc = Arc::new(tx.clone());
let tx_arc2 = tx_arc.clone();
let response = call_stream(
&messages,
&model_name,
&config,
temperature,
max_tokens,
if tools_enabled { Some(&tools) } else { None },
None,
Arc::new(move |delta| {
let content = delta.to_string();
let fut = on_chunk_cb(AiStreamChunk {
content,
done: false,
chunk_type: AiChunkType::Answer,
metadata: None,
children_id: None,
});
fut
}),
Arc::new(move |delta| {
let fut = on_chunk_cb2(AiStreamChunk {
content: delta.to_string(),
done: false,
chunk_type: AiChunkType::Thinking,
metadata: None,
children_id: None,
});
fut
}),
Arc::new(move |tc: &StreamedToolCall| {
let tx = tx_arc2.clone();
let tc_owned = tc.clone();
Box::pin(async move {
let _ = tx.send(tc_owned);
}) as Pin<Box<dyn std::future::Future<Output = ()> + Send>>
}),
)
.await?;
total_input_tokens += response.input_tokens;
total_output_tokens += response.output_tokens;
all_chunks.extend(response.chunks.clone());
let has_tool_calls = tools_enabled && !response.tool_calls.is_empty();
if !has_tool_calls {
return handle_final_answer(
response,
all_chunks,
&request,
session_id,
version_id,
total_input_tokens,
total_output_tokens,
session_start,
)
.await;
}
full_content.push_str(&response.content);
let tool_calls: Vec<ToolCall> = response
.tool_calls
.iter()
.map(|tc| ToolCall {
id: tc.id.clone(),
type_: "function".into(),
function: crate::client::types::ToolCallFunction {
name: tc.name.clone(),
arguments: tc.arguments.clone(),
},
})
.collect();
messages.push(ChatRequestMessage::assistant(
Some(response.content.clone()),
Some(tool_calls.clone()),
));
drain_tool_call_notifications(&mut rx, &on_chunk, &mut all_chunks).await;
let calls: Vec<AgentToolCall> = response
.tool_calls
.iter()
.map(|tc| AgentToolCall {
id: tc.id.clone(),
name: tc.name.clone(),
arguments: tc.arguments.clone(),
})
.collect();
let tool_messages = execute_streaming_tools(
&request,
&calls,
session_id,
&on_chunk,
&mut all_chunks,
tool_registry,
message_builder,
)
.await;
messages.extend(tool_messages);
inject_passive_skills_stream(
&request,
message_builder,
&response.tool_calls,
&mut messages,
)
.await;
tool_depth += 1;
if tool_depth >= max_tool_depth {
let max_depth_text = format!(
"[AI reached maximum tool depth ({}) — no final answer produced]",
max_tool_depth
);
on_chunk(AiStreamChunk {
content: max_depth_text.clone(),
done: true,
chunk_type: AiChunkType::Answer,
metadata: None,
children_id: None,
})
.await;
all_chunks.push(StreamChunk {
chunk_type: StreamChunkType::Answer,
content: max_depth_text,
});
record_ai_session(
&request.cache,
&request.db,
request.project.id,
request.sender.uid,
session_id,
request.room.id,
request.model.id,
version_id.unwrap_or_default(),
total_input_tokens,
total_output_tokens,
session_start.elapsed().as_millis() as i64,
)
.await;
return Ok(StreamResult {
content: full_content,
reasoning_content: String::new(),
input_tokens: 0,
output_tokens: 0,
chunks: all_chunks,
});
}
}
}
async fn drain_tool_call_notifications(
rx: &mut tokio::sync::mpsc::UnboundedReceiver<StreamedToolCall>,
on_chunk: &SharedCallback,
all_chunks: &mut Vec<StreamChunk>,
) {
loop {
match rx.try_recv() {
Ok(tc) => {
let args_display = if tc.arguments.len() > 100 {
let end = tc
.arguments
.char_indices()
.map(|(i, _)| i)
.take_while(|&i| i <= 100)
.last()
.unwrap_or(100);
format!("{}...", &tc.arguments[..end])
} else {
tc.arguments.clone()
};
let tool_display = format!("🔧 {}({})", tc.name, args_display);
// Parse arguments JSON for structured metadata
let args_json =
serde_json::from_str(&tc.arguments).unwrap_or(serde_json::json!({}));
let metadata = serde_json::json!({
"tool": tc.name,
"args": args_json,
"display": tool_display.clone(),
});
on_chunk(AiStreamChunk {
content: tool_display.clone(),
done: false,
chunk_type: AiChunkType::ToolCall,
metadata: Some(metadata),
children_id: None,
})
.await;
all_chunks.push(StreamChunk {
chunk_type: StreamChunkType::ToolCall,
content: tool_display,
});
}
Err(tokio::sync::mpsc::error::TryRecvError::Empty) => break,
Err(tokio::sync::mpsc::error::TryRecvError::Disconnected) => break,
}
}
}
async fn execute_streaming_tools(
request: &AiRequest,
calls: &[AgentToolCall],
session_id: Uuid,
on_chunk: &SharedCallback,
all_chunks: &mut Vec<StreamChunk>,
tool_registry: &Option<crate::tool::registry::ToolRegistry>,
message_builder: &MessageBuilder,
) -> Vec<ChatRequestMessage> {
let mut tool_messages = Vec::new();
let mut ctx = ToolContext::new(
request.db.clone(),
request.cache.clone(),
request.config.clone(),
request.room.id,
Some(request.sender.uid),
)
.with_project(request.project.id);
if let Some(es) = &message_builder.embed_service {
ctx = ctx.with_embed_service(es.clone());
}
if let Some(registry) = tool_registry {
ctx.registry_mut().merge(registry.clone());
}
let recorder =
crate::tool::recorder::ToolCallRecorder::with_session(request.db.clone(), session_id);
let mut join_set = tokio::task::JoinSet::new();
for call in calls {
let call_clone = call.clone();
let mut ctx_clone = ctx.clone();
let sender_uid = request.sender.uid;
let recorder_clone = recorder.clone();
join_set.spawn(async move {
let start = std::time::Instant::now();
let executor = ToolExecutor::new();
let res = executor
.execute_batch(vec![call_clone.clone()], &mut ctx_clone)
.await;
(call_clone, res, start.elapsed(), sender_uid, recorder_clone)
});
}
let heartbeat_dur = std::time::Duration::from_secs(10);
while !join_set.is_empty() {
tokio::select! {
Some(res) = join_set.join_next() => {
if let Ok((call, results, elapsed, sender_uid, recorder)) = res {
match results {
Ok(results) => {
for result in &results {
let text = match &result.result { crate::tool::ToolResult::Ok(v) => v.to_string(), crate::tool::ToolResult::Error(msg) => msg.clone() };
let preview = if text.len() > 300 {
let end = text.char_indices().map(|(i, _)| i).take_while(|&i| i <= 300).last().unwrap_or(300);
format!("{}...", &text[..end])
} else { text.clone() };
tracing::debug!("tool_result: {} — {}", call.name, preview);
let is_error = matches!(result.result, crate::tool::ToolResult::Error(_));
let error_msg = match &result.result { crate::tool::ToolResult::Error(msg) => Some(msg.clone()), _ => None };
recorder.record(crate::tool::recorder::ToolCallRecord {
tool_call_id: call.id.clone(),
session_id: recorder.session_id(),
tool_name: call.name.clone(),
caller: sender_uid,
arguments: call.arguments_json().unwrap_or_default(),
status: if is_error { models::ai::ToolCallStatus::Failed } else { models::ai::ToolCallStatus::Success },
execution_time_ms: Some(elapsed.as_millis() as i64),
error_message: error_msg,
error_stack: None,
retry_count: 0
});
}
let success_display = format!("✅ {}", call.name);
let result_preview: Vec<String> = results.iter().map(|r| {
match &r.result { crate::tool::ToolResult::Ok(v) => v.to_string(), crate::tool::ToolResult::Error(msg) => msg.clone() }
}).collect();
let metadata = serde_json::json!({
"tool": call.name,
"status": "ok",
"result": result_preview.join("\n").chars().take(500).collect::<String>(),
"display": success_display.clone(),
});
on_chunk(AiStreamChunk {
content: success_display.clone(),
done: false,
chunk_type: AiChunkType::ToolResult,
metadata: Some(metadata),
children_id: Some(call.id.clone()),
}).await;
all_chunks.push(StreamChunk { chunk_type: StreamChunkType::ToolCall, content: success_display });
let msgs = ToolExecutor::to_tool_messages(&results);
tool_messages.extend(msgs);
}
Err(e) => {
recorder.record(crate::tool::recorder::ToolCallRecord {
tool_call_id: call.id.clone(),
session_id: recorder.session_id(),
tool_name: call.name.clone(),
caller: sender_uid,
arguments: call.arguments_json().unwrap_or_default(),
status: models::ai::ToolCallStatus::Failed,
execution_time_ms: Some(elapsed.as_millis() as i64),
error_message: Some(e.to_string()),
error_stack: None,
retry_count: 0
});
let err_text = format!("[Tool call failed: {}]", e);
tracing::warn!(tool = %call.name, args = %call.arguments, error = %e, "tool_call_failed");
let err_display = format!("❌ {} (failed)", call.name);
let metadata = serde_json::json!({
"tool": call.name,
"status": "error",
"result": e.to_string(),
"display": err_display.clone(),
});
on_chunk(AiStreamChunk {
content: err_display.clone(),
done: false,
chunk_type: AiChunkType::ToolResult,
metadata: Some(metadata),
children_id: None,
}).await;
all_chunks.push(StreamChunk { chunk_type: StreamChunkType::ToolCall, content: err_display });
tool_messages.push(ChatRequestMessage::tool(&call.id, &err_text));
}
}
}
},
_ = tokio::time::sleep(heartbeat_dur) => {
on_chunk(AiStreamChunk { content: String::new(), done: false, chunk_type: AiChunkType::ToolCall, metadata: None, children_id: None }).await;
}
}
}
tool_messages
}
async fn handle_final_answer(
response: crate::client::StreamResponse,
all_chunks: Vec<StreamChunk>,
request: &AiRequest,
session_id: Uuid,
version_id: Option<Uuid>,
total_input_tokens: i64,
total_output_tokens: i64,
session_start: std::time::Instant,
) -> Result<StreamResult> {
let full_content = response.content.clone();
// Don't push full content as a chunk — incremental deltas in
// response.chunks (already accumulated above) sum to the same text.
// merge_consecutive_blocks would concatenate delta_sum + full =
// 2× full, causing duplicate content in DB persistence.
record_ai_session(
&request.cache,
&request.db,
request.project.id,
request.sender.uid,
session_id,
request.room.id,
request.model.id,
version_id.unwrap_or_default(),
total_input_tokens,
total_output_tokens,
session_start.elapsed().as_millis() as i64,
)
.await;
Ok(StreamResult {
content: full_content,
reasoning_content: response.reasoning_content,
input_tokens: total_input_tokens,
output_tokens: total_output_tokens,
chunks: all_chunks,
})
}
async fn inject_passive_skills_stream(
request: &AiRequest,
message_builder: &MessageBuilder,
tool_calls: &[StreamedToolCall],
messages: &mut Vec<ChatRequestMessage>,
) {
if let Ok(skills) = project_skill::Entity::find()
.filter(project_skill::Column::ProjectUuid.eq(request.project.id))
.filter(project_skill::Column::Enabled.eq(true))
.all(&request.db)
.await
{
let mut skill_entries: Vec<SkillEntry> = skills
.into_iter()
.map(|s| SkillEntry {
slug: s.slug,
name: s.name,
description: s.description,
content: s.content,
})
.collect();
for built_in in crate::skills::all_skills() {
if !skill_entries.iter().any(|s| s.slug == built_in.slug) {
skill_entries.push(SkillEntry {
slug: built_in.slug.to_string(),
name: built_in.name.to_string(),
description: Some(built_in.description.to_string()),
content: built_in.content.clone(),
});
}
}
let tool_events: Vec<ToolCallEvent> = tool_calls
.iter()
.map(|tc| ToolCallEvent {
tool_name: tc.name.clone(),
arguments: tc.arguments.clone(),
})
.collect();
let mut contexts = Vec::new();
for event in &tool_events {
if let Some(ctx) = message_builder
.perception_service
.passive
.detect(event, &skill_entries)
{
MessageBuilder::push_unique_skill_context(&mut contexts, ctx);
}
}
for ctx in contexts {
messages.push(ctx.to_system_message());
}
}
}

831
libs/agent/client/mod.rs
View File

@ -1,831 +0,0 @@
//! Unified AI client with built-in retry, token tracking, and session recording.
//!
//! Uses rig-core as the underlying AI provider library.
pub mod types;
pub use types::{ChatRequestMessage, ToolCall as ClientToolCall};
use std::pin::Pin;
use std::sync::Arc;
use std::time::Instant;
use uuid::Uuid;
use crate::error::{AgentError, Result};
use futures::StreamExt;
use rig::completion::message::{AssistantContent, Message as RigMessage};
use rig::completion::{CompletionModel, GetTokenUsage, ToolDefinition};
use rig::one_or_many::OneOrMany;
use rig::prelude::CompletionClient;
use rig::providers::openai;
/// AI call metrics — increments metrics crate counters for all AI calls.
#[derive(Debug, Clone, Default)]
pub struct AiMetrics;
impl AiMetrics {
pub fn new() -> Self {
Self
}
pub fn record_success(&self, input_tokens: i64, output_tokens: i64, has_function_call: bool) {
metrics::counter!("ai_calls_total").increment(1);
metrics::counter!("ai_calls_success").increment(1);
if input_tokens > 0 {
metrics::counter!("ai_input_tokens_total").increment(input_tokens as u64);
}
if output_tokens > 0 {
metrics::counter!("ai_output_tokens_total").increment(output_tokens as u64);
}
if has_function_call {
metrics::counter!("ai_function_calls_total").increment(1);
}
}
pub fn record_failure(&self) {
metrics::counter!("ai_calls_total").increment(1);
metrics::counter!("ai_calls_failure").increment(1);
}
}
/// Configuration for the AI client.
#[derive(Clone)]
pub struct AiClientConfig {
pub api_key: String,
pub base_url: Option<String>,
}
impl AiClientConfig {
pub fn new(api_key: String) -> Self {
Self {
api_key,
base_url: None,
}
}
pub fn with_base_url(mut self, base_url: impl Into<String>) -> Self {
self.base_url = Some(base_url.into());
self
}
/// Build a rig OpenAI client from this config.
pub fn build_rig_client(&self) -> openai::Client {
let base = self
.base_url
.clone()
.unwrap_or_else(|| "https://api.openai.com".to_string());
openai::Client::builder()
.api_key(&self.api_key)
.base_url(&base)
.build()
.expect("Failed to build rig OpenAI client")
}
}
/// Response from an AI call, including usage statistics.
#[derive(Debug, Clone)]
pub struct AiCallResponse {
pub content: String,
pub input_tokens: i64,
pub output_tokens: i64,
pub latency_ms: i64,
pub tool_calls: Vec<ClientToolCall>,
pub tool_calls_finished: Vec<String>,
}
impl AiCallResponse {
pub fn total_tokens(&self) -> i64 {
self.input_tokens + self.output_tokens
}
}
/// Internal state for retry tracking.
#[derive(Debug)]
struct RetryState {
attempt: u32,
max_retries: u32,
max_backoff_ms: u64,
}
impl RetryState {
fn new(max_retries: u32) -> Self {
Self {
attempt: 0,
max_retries,
max_backoff_ms: 5000,
}
}
fn should_retry(&self) -> bool {
self.attempt < self.max_retries
}
fn backoff_duration(&self) -> std::time::Duration {
let exp = self.attempt.min(5);
let base_ms = 500u64
.saturating_mul(2u64.pow(exp))
.min(self.max_backoff_ms);
let max_jitter = (base_ms / 2).max(base_ms);
let offset = fastrand_u64(max_jitter + 1).saturating_sub(base_ms / 2);
let total = base_ms.saturating_add(offset).min(self.max_backoff_ms);
std::time::Duration::from_millis(total)
}
fn next(&mut self) {
self.attempt += 1;
}
}
fn fastrand_u64(n: u64) -> u64 {
use std::sync::atomic::{AtomicU64, Ordering};
static STATE: AtomicU64 = AtomicU64::new(0x193_667_6a_5e_7c_57);
if n <= 1 {
return 0;
}
let mut current = STATE.load(Ordering::Relaxed);
loop {
let new_val = current.wrapping_mul(6364136223846793005).wrapping_add(1);
match STATE.compare_exchange_weak(current, new_val, Ordering::Relaxed, Ordering::Relaxed) {
Ok(_) => return new_val % n,
Err(actual) => current = actual,
}
}
}
fn is_retryable_error(err: &AgentError) -> bool {
let msg = err.to_string();
msg.contains("connection refused")
|| msg.contains("connection timed out")
|| msg.contains("network error")
|| msg.contains("dns error")
|| msg.contains("error sending request")
|| msg.contains("Http client error")
|| msg.contains("rate_limit")
|| msg.contains("rate limit")
|| msg.contains("429")
|| msg.contains("500")
|| msg.contains("502")
|| msg.contains("503")
|| msg.contains("504")
|| msg.contains("internal_server_error")
|| msg.contains("service_unavailable")
|| msg.contains("gateway_timeout")
|| msg.contains("bad_gateway")
}
static AI_METRICS: std::sync::OnceLock<AiMetrics> = std::sync::OnceLock::new();
fn ai_metrics() -> &'static AiMetrics {
AI_METRICS.get_or_init(AiMetrics::new)
}
// ── Type conversions ─────────────────────────────────────────────────────────
pub(crate) fn to_rig_message(msg: &ChatRequestMessage) -> RigMessage {
match msg.role.as_str() {
"system" => {
// System messages are handled via preamble(), not passed as messages.
// We still need to return a valid RigMessage variant.
RigMessage::user(msg.content.as_deref().unwrap_or(""))
}
"user" => RigMessage::user(msg.content.as_deref().unwrap_or("")),
"assistant" => {
let mut parts: Vec<AssistantContent> = Vec::new();
if let Some(ref content) = msg.content {
if !content.is_empty() {
parts.push(AssistantContent::text(content));
}
}
if let Some(ref tool_calls) = msg.tool_calls {
for tc in tool_calls {
// GLM may return empty tool call IDs — fall back to a generated UUID.
let id = if tc.id.is_empty() {
Uuid::new_v4().to_string()
} else {
tc.id.clone()
};
parts.push(AssistantContent::tool_call_with_call_id(
&id,
id.clone(),
&tc.function.name,
serde_json::from_str(&tc.function.arguments)
.unwrap_or(serde_json::Value::Null),
));
}
}
if parts.is_empty() {
RigMessage::assistant("")
} else if parts.len() == 1 {
// Single part — use simpler constructors
match parts.pop().unwrap() {
AssistantContent::Text(t) => RigMessage::assistant(t.text),
ac => RigMessage::Assistant {
id: None,
content: OneOrMany::one(ac),
},
}
} else {
let content = OneOrMany::many(parts).expect("non-empty parts");
RigMessage::Assistant { id: None, content }
}
}
"tool" | "function" => {
let id = msg.tool_call_id.as_deref().unwrap_or("unknown").to_string();
let call_id = msg.tool_call_id.clone().or_else(|| Some(id.clone()));
let content = msg.content.as_deref().unwrap_or("");
RigMessage::tool_result_with_call_id(id, call_id, content)
}
"developer" => {
// Developer role maps to user/system in rig
RigMessage::user(msg.content.as_deref().unwrap_or(""))
}
_ => RigMessage::user(msg.content.as_deref().unwrap_or("")),
}
}
fn to_rig_tool_def(tool_json: &serde_json::Value) -> Option<ToolDefinition> {
let name = tool_json
.get("function")
.and_then(|f| f.get("name"))
.and_then(|n| n.as_str())?
.to_string();
let description = tool_json
.get("function")
.and_then(|f| f.get("description"))
.and_then(|d| d.as_str())
.map(|s| s.to_string())
.unwrap_or_default();
let parameters = tool_json
.get("function")
.and_then(|f| f.get("parameters"))
.cloned()
.unwrap_or(serde_json::json!({}));
Some(ToolDefinition {
name,
description,
parameters,
})
}
// ── Call helpers ─────────────────────────────────────────────────────────────
async fn do_completion<M>(
model: &M,
messages: &[ChatRequestMessage],
temperature: Option<f64>,
max_tokens: Option<u32>,
tools: Option<&[serde_json::Value]>,
tool_choice: Option<&str>,
) -> Result<(String, u64, u64, Vec<ClientToolCall>, Vec<String>)>
where
M: CompletionModel<Client = openai::Client>,
{
let preamble = messages
.iter()
.find(|m| m.role == "system")
.and_then(|m| m.content.as_deref())
.unwrap_or("")
.to_string();
let non_system: Vec<RigMessage> = messages
.iter()
.filter(|m| m.role != "system")
.map(to_rig_message)
.collect();
let tool_defs: Vec<ToolDefinition> = tools
.map(|ts| ts.iter().filter_map(to_rig_tool_def).collect())
.unwrap_or_default();
let mut builder = model.completion_request("");
if !preamble.is_empty() {
builder = builder.preamble(preamble);
}
if !non_system.is_empty() {
builder = builder.messages(non_system);
}
if let Some(t) = temperature {
builder = builder.temperature(t);
}
if let Some(mt) = max_tokens {
builder = builder.max_tokens(mt as u64);
}
if !tool_defs.is_empty() {
builder = builder.tools(tool_defs);
}
// Only set tool_choice when explicitly provided (mirrors call_stream_once logic)
if let Some(tc) = tool_choice {
match tc {
"none" => {
builder = builder.tool_choice(rig::completion::message::ToolChoice::None);
}
"auto" => {
builder = builder.tool_choice(rig::completion::message::ToolChoice::Auto);
}
s => {
builder = builder.tool_choice(rig::completion::message::ToolChoice::Specific {
function_names: vec![s.to_string()],
});
}
}
}
let response = builder
.send()
.await
.map_err(|e| AgentError::OpenAi(e.to_string()))?;
let mut content = String::new();
let mut tool_names: Vec<String> = Vec::new();
let mut tool_calls: Vec<ClientToolCall> = Vec::new();
for item in response.choice {
match item {
AssistantContent::Text(t) => {
content.push_str(&t.text);
}
AssistantContent::ToolCall(tc) => {
tool_names.push(tc.function.name.clone());
tool_calls.push(ClientToolCall {
id: tc.id,
type_: "function".into(),
function: types::ToolCallFunction {
name: tc.function.name,
arguments: serde_json::to_string(&tc.function.arguments)
.unwrap_or_else(|_| "{}".to_string()),
},
});
}
AssistantContent::Reasoning(_) => {}
AssistantContent::Image(_) => {}
}
}
let input_tokens = response.usage.input_tokens;
let output_tokens = response.usage.output_tokens;
Ok((content, input_tokens, output_tokens, tool_calls, tool_names))
}
// ── Public API ───────────────────────────────────────────────────────────────
/// Call the AI model with automatic retry (no custom params).
pub async fn call_with_retry(
messages: &[ChatRequestMessage],
model_name: &str,
config: &AiClientConfig,
max_retries: Option<u32>,
) -> Result<AiCallResponse> {
let client = config.build_rig_client();
let model = client.completion_model(model_name);
let mut state = RetryState::new(max_retries.unwrap_or(3));
loop {
let start = Instant::now();
let result = do_completion(&model, messages, None, None, None, None).await;
match result {
Ok((content, input_tokens, output_tokens, tool_calls, tool_names)) => {
let latency_ms = start.elapsed().as_millis() as i64;
let has_function_call = !tool_names.is_empty();
ai_metrics().record_success(
input_tokens as i64,
output_tokens as i64,
has_function_call,
);
return Ok(AiCallResponse {
content,
input_tokens: input_tokens as i64,
output_tokens: output_tokens as i64,
latency_ms,
tool_calls,
tool_calls_finished: tool_names,
});
}
Err(ref err) if state.should_retry() && is_retryable_error(err) => {
let duration = state.backoff_duration();
tracing::warn!(
attempt = state.attempt + 1,
max_retries = state.max_retries,
backoff_ms = duration.as_millis() as u64,
model = %model_name,
error = %err,
"ai_call_retry"
);
tokio::time::sleep(duration).await;
state.next();
}
Err(err) => {
ai_metrics().record_failure();
return Err(err);
}
}
}
}
/// Call with custom parameters (temperature, max_tokens, optional tools, optional tool_choice).
pub async fn call_with_params(
messages: &[ChatRequestMessage],
model_name: &str,
config: &AiClientConfig,
temperature: f32,
max_tokens: u32,
max_retries: Option<u32>,
tools: Option<&[serde_json::Value]>,
tool_choice: Option<&str>,
) -> Result<AiCallResponse> {
let client = config.build_rig_client();
let model = client.completion_model(model_name);
let mut state = RetryState::new(max_retries.unwrap_or(3));
loop {
let start = Instant::now();
let result = do_completion(
&model,
messages,
Some(temperature as f64),
Some(max_tokens),
tools,
tool_choice,
)
.await;
match result {
Ok((content, input_tokens, output_tokens, tool_calls, tool_names)) => {
let latency_ms = start.elapsed().as_millis() as i64;
let has_function_call = !tool_names.is_empty();
ai_metrics().record_success(
input_tokens as i64,
output_tokens as i64,
has_function_call,
);
return Ok(AiCallResponse {
content,
input_tokens: input_tokens as i64,
output_tokens: output_tokens as i64,
latency_ms,
tool_calls,
tool_calls_finished: tool_names,
});
}
Err(ref err) if state.should_retry() && is_retryable_error(err) => {
let duration = state.backoff_duration();
tracing::warn!(
attempt = state.attempt + 1,
max_retries = state.max_retries,
backoff_ms = duration.as_millis() as u64,
model = %model_name,
error = %err,
"ai_call_retry"
);
tokio::time::sleep(duration).await;
state.next();
}
Err(err) => {
ai_metrics().record_failure();
return Err(err);
}
}
}
}
/// A tool call extracted from streaming response with accumulated arguments.
#[derive(Debug, Clone)]
pub struct StreamedToolCall {
/// Tool call ID
pub id: String,
/// Tool function name
pub name: String,
/// Accumulated JSON arguments string
pub arguments: String,
}
/// Type of chunk in the streaming response, preserving arrival order.
#[derive(Debug, Clone, PartialEq, Eq)]
pub enum StreamChunkType {
Thinking,
Answer,
ToolCall,
ToolResult,
}
/// A single chunk from the streaming response in arrival order.
#[derive(Debug, Clone)]
pub struct StreamChunk {
pub chunk_type: StreamChunkType,
pub content: String,
}
/// Streaming result from rig.
#[derive(Debug)]
pub struct StreamResponse {
pub content: String,
pub input_tokens: i64,
pub output_tokens: i64,
/// Accumulated reasoning/thinking text from the model.
pub reasoning_content: String,
/// Full tool calls with accumulated arguments (not just names)
pub tool_calls: Vec<StreamedToolCall>,
/// All chunks in arrival order — preserves think/answer/tool interleaving.
pub chunks: Vec<StreamChunk>,
}
/// Async callback: takes a string delta and broadcasts it to the WebSocket.
/// The returned Future must be awaited by the caller.
pub type StreamTextCb =
Arc<dyn Fn(&str) -> Pin<Box<dyn std::future::Future<Output = ()> + Send>> + Send + Sync>;
pub type StreamReasoningCb =
Arc<dyn Fn(&str) -> Pin<Box<dyn std::future::Future<Output = ()> + Send>> + Send + Sync>;
pub type StreamToolCallCb = Arc<
dyn Fn(&StreamedToolCall) -> Pin<Box<dyn std::future::Future<Output = ()> + Send>>
+ Send
+ Sync,
>;
/// Run a streaming chat completion with 60s timeout and 5 retries.
pub async fn call_stream(
messages: &[ChatRequestMessage],
model_name: &str,
config: &AiClientConfig,
temperature: f32,
max_tokens: u32,
tools: Option<&[serde_json::Value]>,
tool_choice: Option<&str>,
on_text_delta: StreamTextCb,
on_reasoning_delta: StreamReasoningCb,
on_tool_call: StreamToolCallCb,
) -> Result<StreamResponse> {
let mut state = RetryState::new(5);
loop {
let result = call_stream_once(
messages,
model_name,
config,
temperature,
max_tokens,
tools,
tool_choice,
on_text_delta.clone(),
on_reasoning_delta.clone(),
on_tool_call.clone(),
)
.await;
match result {
Ok(response) => return Ok(response),
Err(ref err) if state.should_retry() && is_retryable_error(err) => {
let duration = state.backoff_duration();
tracing::warn!(
attempt = state.attempt + 1,
max_retries = 5,
backoff_ms = duration.as_millis() as u64,
model = %model_name,
error = %err,
"ai_stream_retry"
);
tokio::time::sleep(duration).await;
state.next();
}
Err(err) => {
ai_metrics().record_failure();
return Err(err);
}
}
}
}
/// Single attempt of streaming completion with 60s timeout.
async fn call_stream_once(
messages: &[ChatRequestMessage],
model_name: &str,
config: &AiClientConfig,
temperature: f32,
max_tokens: u32,
tools: Option<&[serde_json::Value]>,
tool_choice: Option<&str>,
on_text_delta: StreamTextCb,
on_reasoning_delta: StreamReasoningCb,
on_tool_call: StreamToolCallCb,
) -> Result<StreamResponse> {
let client = config.build_rig_client();
let model = client.completion_model(model_name);
let preamble = messages
.iter()
.find(|m| m.role == "system")
.and_then(|m| m.content.as_deref())
.unwrap_or("")
.to_string();
let non_system: Vec<RigMessage> = messages
.iter()
.filter(|m| m.role != "system")
.map(to_rig_message)
.collect();
let tool_defs: Vec<ToolDefinition> = tools
.map(|ts| ts.iter().filter_map(to_rig_tool_def).collect())
.unwrap_or_default();
let mut builder = model
.completion_request("")
.temperature(temperature as f64)
.max_tokens(max_tokens as u64);
if !preamble.is_empty() {
builder = builder.preamble(preamble);
}
if !non_system.is_empty() {
builder = builder.messages(non_system);
}
if !tool_defs.is_empty() {
builder = builder.tools(tool_defs);
}
if let Some(tc) = tool_choice {
match tc {
"none" => {
builder = builder.tool_choice(rig::completion::message::ToolChoice::None);
}
"auto" => {
builder = builder.tool_choice(rig::completion::message::ToolChoice::Auto);
}
s => {
builder = builder.tool_choice(rig::completion::message::ToolChoice::Specific {
function_names: vec![s.to_string()],
});
}
}
}
let stream_fut = async {
let mut stream = builder
.stream()
.await
.map_err(|e| AgentError::OpenAi(e.to_string()))?;
let mut content = String::new();
let mut reasoning_content = String::new();
let mut tool_calls: Vec<StreamedToolCall> = Vec::new();
let mut chunks: Vec<StreamChunk> = Vec::new();
// Some models (e.g. GLM) ignore tool_choice="none" and still emit tool_calls.
// Filter them out so they don't cause spurious tool execution attempts.
let skip_tool_calls = tool_choice == Some("none");
use std::collections::HashMap;
let mut partial_tool_calls: HashMap<String, StreamedToolCall> = HashMap::new();
let mut stream_finished = false;
use rig::streaming::StreamedAssistantContent;
while let Some(item) = stream.next().await {
match item {
Ok(StreamedAssistantContent::Text(text)) => {
content.push_str(&text.text);
on_text_delta(&text.text).await;
chunks.push(StreamChunk {
chunk_type: StreamChunkType::Answer,
content: text.text,
});
}
Ok(StreamedAssistantContent::ToolCall {
tool_call,
internal_call_id,
}) => {
if skip_tool_calls {
partial_tool_calls.remove(&internal_call_id);
continue;
}
let arguments = match &tool_call.function.arguments {
serde_json::Value::String(s) => s.clone(),
other => serde_json::to_string(other).unwrap_or_else(|_| "{}".to_string()),
};
let tc = StreamedToolCall {
id: tool_call.id.clone(),
name: tool_call.function.name.clone(),
arguments,
};
on_tool_call(&tc).await;
chunks.push(StreamChunk {
chunk_type: StreamChunkType::ToolCall,
content: serde_json::json!({
"id": tc.id,
"name": tc.name,
"arguments": tc.arguments,
})
.to_string(),
});
tool_calls.push(tc);
partial_tool_calls.remove(&internal_call_id);
}
Ok(StreamedAssistantContent::ToolCallDelta {
id,
internal_call_id,
content: delta_content,
}) => {
if skip_tool_calls {
continue;
}
use rig::streaming::ToolCallDeltaContent;
match delta_content {
ToolCallDeltaContent::Name(name) => {
partial_tool_calls.insert(
internal_call_id.clone(),
StreamedToolCall {
id: id.clone(),
name,
arguments: String::new(),
},
);
}
ToolCallDeltaContent::Delta(delta) => {
if let Some(tc) = partial_tool_calls.get_mut(&internal_call_id) {
tc.arguments.push_str(&delta);
}
}
}
}
Ok(StreamedAssistantContent::Reasoning(reasoning)) => {
for part in &reasoning.content {
if let rig::completion::message::ReasoningContent::Text { text, .. } = part
{
reasoning_content.push_str(text);
on_reasoning_delta(text).await;
chunks.push(StreamChunk {
chunk_type: StreamChunkType::Thinking,
content: text.clone(),
});
}
}
}
Ok(StreamedAssistantContent::ReasoningDelta { reasoning, .. }) => {
reasoning_content.push_str(&reasoning);
on_reasoning_delta(&reasoning).await;
chunks.push(StreamChunk {
chunk_type: StreamChunkType::Thinking,
content: reasoning.clone(),
});
}
Ok(StreamedAssistantContent::Final(response)) => {
stream_finished = true;
if !skip_tool_calls {
for (_, tc) in partial_tool_calls.drain() {
tool_calls.push(tc);
}
} else {
partial_tool_calls.drain();
}
if let Some(usage) = response.token_usage() {
let in_toks = usage.input_tokens as i64;
let out_toks = usage.output_tokens as i64;
ai_metrics().record_success(in_toks, out_toks, !tool_calls.is_empty());
return Ok(StreamResponse {
content,
reasoning_content,
input_tokens: in_toks,
output_tokens: out_toks,
tool_calls,
chunks,
});
}
// Usage not available from Final — fall through to flush
}
Err(e) => return Err(AgentError::OpenAi(e.to_string())),
}
}
// Flush any remaining partial tool calls (if stream ended without Final or Final had no usage)
if !stream_finished && !skip_tool_calls {
for (_, tc) in partial_tool_calls.drain() {
tool_calls.push(tc);
}
}
ai_metrics().record_success(0, 0, !tool_calls.is_empty());
Ok(StreamResponse {
content,
reasoning_content,
input_tokens: 0,
output_tokens: 0,
tool_calls,
chunks,
})
};
// 120s timeout for the entire stream
match tokio::time::timeout(std::time::Duration::from_secs(120), stream_fut).await {
Ok(result) => result,
Err(_) => Err(AgentError::Timeout {
task_id: 0,
seconds: 120,
}),
}
}

240
libs/agent/client/types.rs
View File

@ -1,240 +0,0 @@
//! Internal message types for OpenAI-compatible chat completion API.
//!
//! Uses plain structs with `role: String` instead of an enum — easier to serialize,
//! and the downstream code only constructs specific variants anyway.
use serde::{Deserialize, Serialize};
/// A message in a chat completion request.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ChatRequestMessage {
/// One of "system", "user", "assistant", "tool", "developer", "function"
pub role: String,
#[serde(skip_serializing_if = "Option::is_none")]
pub content: Option<String>,
#[serde(skip_serializing_if = "Option::is_none")]
pub name: Option<String>,
/// Required for "tool" role messages
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_call_id: Option<String>,
/// Tool calls for "assistant" role messages
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_calls: Option<Vec<ToolCall>>,
}
impl ChatRequestMessage {
pub fn system(content: impl Into<String>) -> Self {
Self {
role: "system".into(),
content: Some(content.into()),
name: None,
tool_call_id: None,
tool_calls: None,
}
}
pub fn user(content: impl Into<String>) -> Self {
Self {
role: "user".into(),
content: Some(content.into()),
name: None,
tool_call_id: None,
tool_calls: None,
}
}
pub fn assistant(content: Option<String>, tool_calls: Option<Vec<ToolCall>>) -> Self {
Self {
role: "assistant".into(),
content,
name: None,
tool_call_id: None,
tool_calls,
}
}
pub fn tool(tool_call_id: impl Into<String>, content: impl Into<String>) -> Self {
Self {
role: "tool".into(),
content: Some(content.into()),
name: None,
tool_call_id: Some(tool_call_id.into()),
tool_calls: None,
}
}
pub fn with_name(mut self, name: impl Into<String>) -> Self {
self.name = Some(name.into());
self
}
pub fn developer(content: impl Into<String>) -> Self {
Self {
role: "developer".into(),
content: Some(content.into()),
name: None,
tool_call_id: None,
tool_calls: None,
}
}
/// Creates a function/assistant message with tool_calls (used to record the AI's tool call).
pub fn with_tool_calls(mut self, tool_calls: Vec<ToolCall>) -> Self {
self.tool_calls = Some(tool_calls);
self
}
}
/// A tool call within an assistant message.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ToolCall {
pub id: String,
#[serde(rename = "type")]
pub type_: String,
pub function: ToolCallFunction,
}
/// Function details within a tool call.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ToolCallFunction {
pub name: String,
pub arguments: String,
}
/// Chat completion request body (serialized to JSON for the HTTP API).
#[derive(Debug, Clone, Serialize)]
pub struct ChatCompletionRequest {
pub model: String,
pub messages: Vec<ChatRequestMessage>,
#[serde(skip_serializing_if = "Option::is_none")]
pub temperature: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub max_completion_tokens: Option<u32>,
#[serde(skip_serializing_if = "Option::is_none")]
pub top_p: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub frequency_penalty: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub presence_penalty: Option<f64>,
#[serde(skip_serializing_if = "Option::is_none")]
pub stream: Option<bool>,
#[serde(skip_serializing_if = "Option::is_none")]
pub reasoning_effort: Option<ReasoningEffort>,
#[serde(skip_serializing_if = "Option::is_none")]
pub tools: Option<Vec<serde_json::Value>>,
#[serde(skip_serializing_if = "Option::is_none")]
pub tool_choice: Option<serde_json::Value>,
}
impl ChatCompletionRequest {
pub fn with_stream(mut self) -> Self {
self.stream = Some(true);
self
}
}
/// Reasoning effort level for supported models.
#[derive(Debug, Clone, Serialize)]
#[serde(rename_all = "lowercase")]
pub enum ReasoningEffort {
High,
}
// ── Response types (non-streaming) ──
/// Chat completion response (non-streaming). Deserialize-only from the API JSON.
#[derive(Debug, Clone, Deserialize)]
pub struct ChatCompletionResponse {
#[serde(default)]
pub id: Option<String>,
#[serde(default)]
pub model: Option<String>,
pub choices: Vec<Choice>,
#[serde(default)]
pub usage: Option<Usage>,
}
#[derive(Debug, Clone, Deserialize)]
pub struct Choice {
pub index: u32,
pub message: ResponseMessage,
pub finish_reason: Option<String>,
}
#[derive(Debug, Clone, Deserialize)]
pub struct ResponseMessage {
pub role: Option<String>,
pub content: Option<String>,
#[serde(default)]
pub tool_calls: Option<Vec<ResponseToolCall>>,
}
#[derive(Debug, Clone, Deserialize)]
pub struct ResponseToolCall {
pub id: String,
#[serde(rename = "type")]
pub type_: String,
pub function: ResponseToolCallFunction,
}
#[derive(Debug, Clone, Deserialize)]
pub struct ResponseToolCallFunction {
pub name: String,
pub arguments: String,
}
/// Token usage from the response.
#[derive(Debug, Clone, Deserialize)]
pub struct Usage {
#[serde(rename = "prompt_tokens", alias = "input_tokens")]
pub prompt_tokens: u64,
#[serde(rename = "completion_tokens", alias = "output_tokens")]
pub completion_tokens: u64,
}
// ── Streaming types ──
/// A chunk from a streaming chat completion (SSE `data:` lines).
#[derive(Debug, Clone, Deserialize)]
pub struct StreamChunk {
#[serde(default)]
pub id: Option<String>,
#[serde(default)]
pub model: Option<String>,
pub choices: Vec<StreamChoice>,
}
#[derive(Debug, Clone, Deserialize)]
pub struct StreamChoice {
pub delta: Delta,
pub finish_reason: Option<String>,
pub index: u32,
}
#[derive(Debug, Clone, Deserialize)]
pub struct Delta {
#[serde(default)]
pub role: Option<String>,
#[serde(default)]
pub content: Option<String>,
#[serde(default)]
pub tool_calls: Option<Vec<StreamToolCall>>,
}
#[derive(Debug, Clone, Deserialize)]
pub struct StreamToolCall {
pub index: u32,
#[serde(default)]
pub id: Option<String>,
#[serde(rename = "type", default)]
pub type_: Option<String>,
pub function: Option<StreamToolCallFunction>,
}
#[derive(Debug, Clone, Deserialize)]
pub struct StreamToolCallFunction {
#[serde(default)]
pub name: Option<String>,
#[serde(default)]
pub arguments: Option<String>,
}

39
libs/agent/compact/auth_fetch.rs
View File

@ -1,39 +0,0 @@
use crate::AgentError;
use models::Expr;
use models::rooms::room_message::{
Column as RmCol, Entity as RoomMessage, Model as RoomMessageModel,
};
use sea_orm::*;
impl super::CompactService {
pub async fn fetch_room_messages_secure(
&self,
room_id: uuid::Uuid,
requester_id: uuid::Uuid,
) -> Result<Vec<RoomMessageModel>, AgentError> {
use models::rooms::{RoomAccess, RoomUserState};
RoomMessage::find()
.filter(RmCol::Room.eq(room_id))
.filter(
Condition::any()
.add(Expr::exists(
RoomUserState::find()
.filter(models::rooms::room_user_state::Column::Room.eq(room_id))
.filter(models::rooms::room_user_state::Column::User.eq(requester_id))
.into_query(),
))
.add(Expr::exists(
RoomAccess::find()
.filter(models::rooms::room_access::Column::Room.eq(room_id))
.filter(models::rooms::room_access::Column::User.eq(requester_id))
.into_query(),
)),
)
.order_by_asc(RmCol::Seq)
.limit(10000)
.all(&self.db)
.await
.map_err(|e| AgentError::Internal(e.to_string()))
}
}

45
libs/agent/compact/helpers.rs
View File

@ -1,45 +0,0 @@
use super::types::{CompactSummary, MessageSummary};
pub fn messages_to_text<F>(
messages: &[models::rooms::room_message::Model],
sender_mapper: F,
) -> String
where
F: Fn(&models::rooms::room_message::Model) -> String,
{
messages
.iter()
.map(|m| {
let sender = sender_mapper(m);
format!("[{}] {}: {}", m.send_at, sender, m.content)
})
.collect::<Vec<_>>()
.join("\n")
}
pub fn retained_as_text(retained: &[MessageSummary]) -> String {
retained
.iter()
.map(|m| format!("[{}] {}: {}", m.send_at, m.sender_name, m.content))
.collect::<Vec<_>>()
.join("\n")
}
pub fn summary_content(summary: &CompactSummary) -> String {
if summary.summary.is_empty() {
format!(
"## Recent conversation ({} messages)\n\n{}",
summary.retained.len(),
retained_as_text(&summary.retained)
)
} else {
format!(
"## Earlier conversation ({} messages summarised)\n{}\n\n\
## Most recent {} messages\n\n{}",
summary.messages_compressed,
summary.summary,
summary.retained.len(),
retained_as_text(&summary.retained)
)
}
}

56
libs/agent/compact/mod.rs
View File

@ -1,56 +0,0 @@
//! Context compaction for AI sessions and room message history.
pub mod auth_fetch;
pub mod helpers;
pub mod room_compactor;
pub mod summarizer;
pub mod types;
use sea_orm::DatabaseConnection;
pub use types::{
CompactConfig, CompactLevel, CompactSummary, MessageSummary, RoomCompactContext,
RoomCompactRecord, ThresholdResult,
};
#[derive(Clone)]
pub struct CompactService {
db: DatabaseConnection,
ai_client_config: crate::client::AiClientConfig,
model: String,
model_context_limit: Option<usize>,
}
impl CompactService {
pub fn new(
db: DatabaseConnection,
ai_client_config: crate::client::AiClientConfig,
model: String,
) -> Self {
Self {
db,
ai_client_config,
model,
model_context_limit: None,
}
}
pub fn for_model(&self, model: impl Into<String>) -> Self {
Self {
db: self.db.clone(),
ai_client_config: self.ai_client_config.clone(),
model: model.into(),
model_context_limit: self.model_context_limit,
}
}
pub fn with_model_context_limit(mut self, model_context_limit: Option<usize>) -> Self {
self.model_context_limit = model_context_limit.filter(|limit| *limit > 0);
self
}
pub fn for_model_entry(&self, model: &models::agents::model::Model) -> Self {
self.for_model(model.name.clone())
.with_model_context_limit(Some(model.context_length.max(0) as usize))
}
}

422
libs/agent/compact/room_compactor.rs
View File

@ -1,422 +0,0 @@
use models::rooms::room_message::{
Column as RmCol, Entity as RoomMessage, Model as RoomMessageModel,
};
use sea_orm::ColumnTrait;
use sea_orm::{ConnectionTrait, EntityTrait, QueryFilter, QueryOrder, QuerySelect};
use crate::compact::types::{CompactConfig, CompactLevel, RoomCompactContext, RoomCompactRecord};
use crate::tokent::resolve_usage;
use crate::{AgentError, CompactSummary, MessageSummary};
impl super::CompactService {
pub async fn latest_room_compact_record(
&self,
room_id: uuid::Uuid,
) -> Result<Option<RoomCompactRecord>, AgentError> {
let stmt = sea_orm::Statement::from_sql_and_values(
sea_orm::DbBackend::Postgres,
"SELECT id, room, from_seq, to_seq, summary, message_count, source_message_ids, created_at \
FROM room_compact_summary WHERE room = $1 ORDER BY to_seq DESC, created_at DESC LIMIT 1",
vec![room_id.into()],
);
let Some(row) = self
.db
.query_one_raw(stmt)
.await
.map_err(|e| AgentError::Internal(e.to_string()))?
else {
return Ok(None);
};
let source_json: serde_json::Value = row
.try_get("", "source_message_ids")
.map_err(|e| AgentError::Internal(e.to_string()))?;
let source_message_ids = source_json
.as_array()
.map(|ids| {
ids.iter()
.filter_map(|v| v.as_str())
.filter_map(|s| uuid::Uuid::parse_str(s).ok())
.collect::<Vec<_>>()
})
.unwrap_or_default();
Ok(Some(RoomCompactRecord {
id: row
.try_get("", "id")
.map_err(|e| AgentError::Internal(e.to_string()))?,
room_id: row
.try_get("", "room")
.map_err(|e| AgentError::Internal(e.to_string()))?,
from_seq: row
.try_get("", "from_seq")
.map_err(|e| AgentError::Internal(e.to_string()))?,
to_seq: row
.try_get("", "to_seq")
.map_err(|e| AgentError::Internal(e.to_string()))?,
summary: row
.try_get("", "summary")
.map_err(|e| AgentError::Internal(e.to_string()))?,
message_count: row
.try_get("", "message_count")
.map_err(|e| AgentError::Internal(e.to_string()))?,
source_message_ids,
created_at: row
.try_get("", "created_at")
.map_err(|e| AgentError::Internal(e.to_string()))?,
}))
}
async fn insert_room_compact_record(
&self,
room_id: uuid::Uuid,
from_seq: i64,
to_seq: i64,
summary: &str,
source_message_ids: &[uuid::Uuid],
) -> Result<RoomCompactRecord, AgentError> {
let id = uuid::Uuid::new_v4();
let now = chrono::Utc::now();
let source_json = serde_json::Value::Array(
source_message_ids
.iter()
.map(|id| serde_json::Value::String(id.to_string()))
.collect(),
);
let stmt = sea_orm::Statement::from_sql_and_values(
sea_orm::DbBackend::Postgres,
"INSERT INTO room_compact_summary \
(id, room, from_seq, to_seq, summary, message_count, source_message_ids, created_at, updated_at) \
VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9)",
vec![
id.into(),
room_id.into(),
from_seq.into(),
to_seq.into(),
summary.to_string().into(),
(source_message_ids.len() as i32).into(),
source_json.into(),
now.into(),
now.into(),
],
);
self.db
.execute_raw(stmt)
.await
.map_err(|e| AgentError::Internal(e.to_string()))?;
Ok(RoomCompactRecord {
id,
room_id,
from_seq,
to_seq,
summary: summary.to_string(),
message_count: source_message_ids.len() as i32,
source_message_ids: source_message_ids.to_vec(),
created_at: now,
})
}
fn clean_dedupe_sort_messages(mut messages: Vec<RoomMessageModel>) -> Vec<RoomMessageModel> {
messages.retain(|m| {
m.revoked.is_none()
&& !m.content.trim().is_empty()
&& matches!(m.content_type, models::rooms::MessageContentType::Text)
});
messages.sort_by_key(|m| (m.seq, m.send_at));
let mut seen = std::collections::HashSet::new();
messages
.into_iter()
.filter(|m| {
let normalized = m
.content
.split_whitespace()
.collect::<Vec<_>>()
.join(" ")
.to_lowercase();
let key = format!("{}:{:?}:{}", m.sender_type, m.sender_id, normalized);
seen.insert(key)
})
.collect()
}
fn resolve_retain_count(config: CompactConfig, estimated_tokens: usize) -> usize {
let level = if config.auto_level {
CompactLevel::auto_select(estimated_tokens, config.token_threshold)
} else {
config.default_level
};
level.retain_count()
}
pub async fn prepare_room_compact_context(
&self,
room_id: uuid::Uuid,
requester_id: uuid::Uuid,
user_names: Option<std::collections::HashMap<uuid::Uuid, String>>,
config: CompactConfig,
) -> Result<RoomCompactContext, AgentError> {
let latest = self.latest_room_compact_record(room_id).await?;
let cutoff_seq = latest.as_ref().map(|r| r.to_seq);
let previous_summary = latest.as_ref().map(|r| r.summary.as_str());
let messages = self
.fetch_room_messages_secure(room_id, requester_id)
.await?;
let messages = messages
.into_iter()
.filter(|m| cutoff_seq.map(|seq| m.seq > seq).unwrap_or(true))
.collect::<Vec<_>>();
let messages = Self::clean_dedupe_sort_messages(messages);
let user_ids: Vec<uuid::Uuid> = messages
.iter()
.filter_map(|m| m.sender_id)
.collect::<std::collections::HashSet<_>>()
.into_iter()
.collect();
let user_name_map = match user_names {
Some(map) => map,
None => self.get_user_name_map(&user_ids).await?,
};
let sender_mapper = |m: &RoomMessageModel| {
if let Some(user_id) = m.sender_id {
if let Some(username) = user_name_map.get(&user_id) {
return username.clone();
}
}
m.sender_type.to_string()
};
let incremental_text = crate::compact::helpers::messages_to_text(&messages, sender_mapper);
let estimate_input = match previous_summary {
Some(summary) if !summary.is_empty() => format!("{}\n{}", summary, incremental_text),
_ => incremental_text.clone(),
};
let estimated_tokens = crate::tokent::count_message_text(&estimate_input, &self.model)
.unwrap_or_else(|_| estimate_input.len() / 4);
let retain_count = Self::resolve_retain_count(config, estimated_tokens);
if estimated_tokens >= config.token_threshold && messages.len() > retain_count {
let split_index = messages.len().saturating_sub(retain_count);
let (to_summarize, retained_messages) = messages.split_at(split_index);
let from_seq = to_summarize
.first()
.map(|m| m.seq)
.unwrap_or(cutoff_seq.unwrap_or(0) + 1);
let to_seq = to_summarize.last().map(|m| m.seq).unwrap_or(from_seq);
let source_ids: Vec<uuid::Uuid> = to_summarize.iter().map(|m| m.id).collect();
let (summary, _usage) = self
.summarize_room_increment(previous_summary, to_summarize, config.max_summary_tokens)
.await?;
let record = self
.insert_room_compact_record(room_id, from_seq, to_seq, &summary, &source_ids)
.await?;
let retained = retained_messages
.iter()
.map(|m| Self::message_to_summary(m, &user_name_map))
.collect();
return Ok(RoomCompactContext {
room_id,
cutoff_seq: Some(record.to_seq),
summary: Some(record.summary),
retained,
estimated_tokens,
compacted: true,
});
}
let retained = messages
.iter()
.rev()
.take(50)
.collect::<Vec<_>>()
.into_iter()
.rev()
.map(|m| Self::message_to_summary(m, &user_name_map))
.collect();
Ok(RoomCompactContext {
room_id,
cutoff_seq,
summary: latest.map(|r| r.summary),
retained,
estimated_tokens,
compacted: false,
})
}
pub async fn compact_room(
&self,
room_id: uuid::Uuid,
level: CompactLevel,
user_names: Option<std::collections::HashMap<uuid::Uuid, String>>,
requester_id: uuid::Uuid,
context_window_tokens: i32,
compaction_max_summary_ratio: f32,
) -> Result<CompactSummary, AgentError> {
let messages = self
.fetch_room_messages_secure(room_id, requester_id)
.await?;
if messages.is_empty() {
let room_exists = models::rooms::room::Entity::find_by_id(room_id)
.one(&self.db)
.await
.map_err(|e| AgentError::Internal(e.to_string()))?
.is_some();
if room_exists {
return Err(AgentError::Internal("Access denied or room empty".into()));
} else {
return Err(AgentError::Internal("Room not found".into()));
}
}
let user_ids: Vec<uuid::Uuid> = messages
.iter()
.filter_map(|m| m.sender_id)
.collect::<std::collections::HashSet<_>>()
.into_iter()
.collect();
let user_name_map = match user_names {
Some(map) => map,
None => self.get_user_name_map(&user_ids).await?,
};
if messages.len() <= level.retain_count() {
let retained: Vec<MessageSummary> = messages
.iter()
.map(|m| Self::message_to_summary(m, &user_name_map))
.collect();
return Ok(CompactSummary {
session_id: uuid::Uuid::new_v4(),
room_id,
retained,
summary: String::new(),
compacted_at: chrono::Utc::now(),
messages_compressed: 0,
usage: None,
});
}
let retain_count = level.retain_count();
let split_index = messages.len().saturating_sub(retain_count);
let (to_summarize, retained_messages) = messages.split_at(split_index);
let retained: Vec<MessageSummary> = retained_messages
.iter()
.map(|m| Self::message_to_summary(m, &user_name_map))
.collect();
let max_summary_tokens = CompactConfig::summary_token_budget(
context_window_tokens.max(0) as usize,
compaction_max_summary_ratio,
);
let (summary, remote_usage) = self
.summarize_messages(to_summarize, max_summary_tokens)
.await?;
let summarized_text = to_summarize
.iter()
.map(|m| m.content.as_str())
.collect::<Vec<_>>()
.join("\n");
let usage = resolve_usage(remote_usage, &self.model, &summarized_text, &summary);
Ok(CompactSummary {
session_id: uuid::Uuid::new_v4(),
room_id,
retained,
summary,
compacted_at: chrono::Utc::now(),
messages_compressed: to_summarize.len(),
usage: Some(usage),
})
}
pub async fn compact_session(
&self,
session_id: uuid::Uuid,
level: CompactLevel,
user_names: Option<std::collections::HashMap<uuid::Uuid, String>>,
context_window_tokens: i32,
compaction_max_summary_ratio: f32,
) -> Result<CompactSummary, AgentError> {
let messages: Vec<RoomMessageModel> = RoomMessage::find()
.filter(RmCol::Room.eq(session_id))
.order_by_asc(RmCol::Seq)
.limit(10000)
.all(&self.db)
.await
.map_err(|e| AgentError::Internal(e.to_string()))?;
if messages.is_empty() {
return Err(AgentError::Internal("session has no messages".into()));
}
let user_ids: Vec<uuid::Uuid> = messages
.iter()
.filter_map(|m| m.sender_id)
.collect::<std::collections::HashSet<_>>()
.into_iter()
.collect();
let user_name_map = match user_names {
Some(map) => map,
None => self.get_user_name_map(&user_ids).await?,
};
if messages.len() <= level.retain_count() {
let retained: Vec<MessageSummary> = messages
.iter()
.map(|m| Self::message_to_summary(m, &user_name_map))
.collect();
return Ok(CompactSummary {
session_id,
room_id: uuid::Uuid::nil(),
retained,
summary: String::new(),
compacted_at: chrono::Utc::now(),
messages_compressed: 0,
usage: None,
});
}
let retain_count = level.retain_count();
let split_index = messages.len().saturating_sub(retain_count);
let (to_summarize, retained_messages) = messages.split_at(split_index);
let retained: Vec<MessageSummary> = retained_messages
.iter()
.map(|m| Self::message_to_summary(m, &user_name_map))
.collect();
let max_summary_tokens = CompactConfig::summary_token_budget(
context_window_tokens.max(0) as usize,
compaction_max_summary_ratio,
);
let (summary, remote_usage) = self
.summarize_messages(to_summarize, max_summary_tokens)
.await?;
let summarized_text = to_summarize
.iter()
.map(|m| m.content.as_str())
.collect::<Vec<_>>()
.join("\n");
let usage = resolve_usage(remote_usage, &self.model, &summarized_text, &summary);
Ok(CompactSummary {
session_id,
room_id: uuid::Uuid::nil(),
retained,
summary,
compacted_at: chrono::Utc::now(),
messages_compressed: to_summarize.len(),
usage: Some(usage),
})
}
}

513
libs/agent/compact/summarizer.rs
View File

@ -1,513 +0,0 @@
use models::rooms::room_message::Model as RoomMessageModel;
use models::users::user::{Column as UserCol, Entity as User};
use sea_orm::{ColumnTrait, EntityTrait, QueryFilter};
use crate::AgentError;
use crate::client::call_with_params;
use crate::client::types::ChatRequestMessage;
use crate::compact::types::{CompactConfig, MessageSummary};
use crate::tokent::{TokenUsage, count_message_text};
const DEFAULT_MODEL_CONTEXT_LIMIT: usize = 128_000;
const MODEL_INPUT_RATIO_NUMERATOR: usize = 85;
const MODEL_INPUT_RATIO_DENOMINATOR: usize = 100;
const MIN_ROUND_SUMMARY_TOKENS: usize = 64;
#[derive(Clone, Copy)]
enum SummaryKind {
Conversation,
RoomIncrement,
}
impl super::CompactService {
pub async fn summarize_room_increment(
&self,
previous_summary: Option<&str>,
messages: &[RoomMessageModel],
max_summary_tokens: usize,
) -> Result<(String, Option<TokenUsage>), AgentError> {
let user_ids: Vec<uuid::Uuid> = messages
.iter()
.filter_map(|m| m.sender_id)
.collect::<std::collections::HashSet<_>>()
.into_iter()
.collect();
let user_name_map = self.get_user_name_map(&user_ids).await?;
let blocks = messages
.iter()
.map(|m| {
let sender = if let Some(user_id) = m.sender_id {
user_name_map
.get(&user_id)
.cloned()
.unwrap_or_else(|| m.sender_type.to_string())
} else {
m.sender_type.to_string()
};
format!("[{}] {}: {}", m.send_at, sender, m.content)
})
.collect::<Vec<_>>();
self.summarize_blocks_with_optional_previous(
blocks,
previous_summary,
max_summary_tokens,
SummaryKind::RoomIncrement,
)
.await
}
pub async fn summarize_messages(
&self,
messages: &[RoomMessageModel],
max_summary_tokens: usize,
) -> Result<(String, Option<TokenUsage>), AgentError> {
let user_ids: Vec<uuid::Uuid> = messages
.iter()
.filter_map(|m| m.sender_id)
.collect::<std::collections::HashSet<_>>()
.into_iter()
.collect();
let user_name_map = self.get_user_name_map(&user_ids).await?;
let blocks = messages
.iter()
.map(|m| {
let sender = if let Some(user_id) = m.sender_id {
user_name_map
.get(&user_id)
.cloned()
.unwrap_or_else(|| m.sender_type.to_string())
} else {
m.sender_type.to_string()
};
format!("[{}] {}: {}", m.send_at, sender, m.content)
})
.collect::<Vec<_>>();
self.summarize_blocks_with_optional_previous(
blocks,
None,
max_summary_tokens,
SummaryKind::Conversation,
)
.await
}
pub fn message_to_summary(
m: &RoomMessageModel,
user_name_map: &std::collections::HashMap<uuid::Uuid, String>,
) -> MessageSummary {
let sender_name = if let Some(user_id) = m.sender_id {
user_name_map
.get(&user_id)
.cloned()
.unwrap_or_else(|| m.sender_type.to_string())
} else {
m.sender_type.to_string()
};
MessageSummary {
id: m.id,
sender_type: m.sender_type.clone(),
sender_id: m.sender_id,
sender_name,
content: m.content.clone(),
content_type: m.content_type.clone(),
tool_call_id: None,
send_at: m.send_at,
}
}
pub async fn get_user_name_map(
&self,
user_ids: &[uuid::Uuid],
) -> Result<std::collections::HashMap<uuid::Uuid, String>, AgentError> {
use std::collections::HashMap;
let mut map = HashMap::new();
if !user_ids.is_empty() {
let users = User::find()
.filter(UserCol::Uid.is_in(user_ids.to_vec()))
.all(&self.db)
.await
.map_err(|e| AgentError::Internal(e.to_string()))?;
for user in users {
map.insert(user.uid, user.username);
}
}
Ok(map)
}
async fn summarize_blocks_with_optional_previous(
&self,
blocks: Vec<String>,
previous_summary: Option<&str>,
max_summary_tokens: usize,
kind: SummaryKind,
) -> Result<(String, Option<TokenUsage>), AgentError> {
let final_budget = Self::final_summary_budget(max_summary_tokens);
let input_budget = self.safe_model_input_budget();
let round_budget = Self::round_summary_budget(final_budget, input_budget);
let mut total_usage = TokenUsage::default();
let mut has_usage = false;
let fitted_chunks =
self.split_blocks_to_fit(blocks, input_budget, round_budget, kind, false)?;
let mut partial_summaries = Vec::new();
for chunk in fitted_chunks {
let prompt = self.build_prompt(kind, false, &chunk, round_budget);
let (summary, usage) = self
.invoke_summary_prompt(&prompt, round_budget, Self::temperature_for(kind))
.await?;
Self::accumulate_usage(&mut total_usage, &mut has_usage, usage);
partial_summaries.push(summary);
}
if let Some(previous) = previous_summary
.map(str::trim)
.filter(|summary| !summary.is_empty())
{
partial_summaries.insert(0, previous.to_string());
}
if partial_summaries.is_empty() {
return Ok((String::new(), None));
}
if partial_summaries.len() == 1 && previous_summary.is_none() {
return Ok((
partial_summaries.remove(0),
if has_usage { Some(total_usage) } else { None },
));
}
let final_summary = self
.merge_summary_rounds(
partial_summaries,
final_budget,
round_budget,
kind,
&mut total_usage,
&mut has_usage,
)
.await?;
Ok((
final_summary,
if has_usage { Some(total_usage) } else { None },
))
}
async fn merge_summary_rounds(
&self,
mut summaries: Vec<String>,
final_budget: usize,
round_budget: usize,
kind: SummaryKind,
total_usage: &mut TokenUsage,
has_usage: &mut bool,
) -> Result<String, AgentError> {
let input_budget = self.safe_model_input_budget();
while summaries.len() > 1 {
let current_budget = if summaries.len() <= 2 {
final_budget
} else {
round_budget
};
let mut next_round = Vec::new();
let mut idx = 0usize;
while idx < summaries.len() {
if idx + 1 >= summaries.len() {
next_round.push(summaries[idx].clone());
idx += 1;
continue;
}
let pair = vec![summaries[idx].clone(), summaries[idx + 1].clone()];
let fitted_pairs =
self.split_blocks_to_fit(pair, input_budget, current_budget, kind, true)?;
for pair_text in fitted_pairs {
let prompt = self.build_prompt(kind, true, &pair_text, current_budget);
let (summary, usage) = self
.invoke_summary_prompt(&prompt, current_budget, Self::temperature_for(kind))
.await?;
Self::accumulate_usage(total_usage, has_usage, usage);
next_round.push(summary);
}
idx += 2;
}
summaries = next_round;
}
summaries
.pop()
.ok_or_else(|| AgentError::Internal("summary merge produced no output".into()))
}
async fn invoke_summary_prompt(
&self,
prompt: &str,
max_summary_tokens: usize,
temperature: f32,
) -> Result<(String, Option<TokenUsage>), AgentError> {
let response = call_with_params(
&[ChatRequestMessage::user(prompt.to_string())],
&self.model,
&self.ai_client_config,
temperature,
max_summary_tokens as u32,
None,
None,
None,
)
.await
.map_err(|e| AgentError::OpenAi(e.to_string()))?;
let usage =
TokenUsage::from_remote(response.input_tokens as u32, response.output_tokens as u32);
Ok((response.content, usage))
}
fn split_blocks_to_fit(
&self,
blocks: Vec<String>,
input_budget: usize,
max_summary_tokens: usize,
kind: SummaryKind,
is_merge: bool,
) -> Result<Vec<String>, AgentError> {
let mut chunks = Vec::new();
self.collect_fitting_chunks(
blocks,
input_budget,
max_summary_tokens,
kind,
is_merge,
&mut chunks,
)?;
Ok(chunks)
}
fn collect_fitting_chunks(
&self,
blocks: Vec<String>,
input_budget: usize,
max_summary_tokens: usize,
kind: SummaryKind,
is_merge: bool,
chunks: &mut Vec<String>,
) -> Result<(), AgentError> {
let body = Self::join_blocks(&blocks, is_merge);
let prompt = self.build_prompt(kind, is_merge, &body, max_summary_tokens);
if self.estimate_tokens(&prompt) <= input_budget {
chunks.push(body);
return Ok(());
}
if blocks.len() > 1 {
let mid = blocks.len() / 2;
self.collect_fitting_chunks(
blocks[..mid].to_vec(),
input_budget,
max_summary_tokens,
kind,
is_merge,
chunks,
)?;
self.collect_fitting_chunks(
blocks[mid..].to_vec(),
input_budget,
max_summary_tokens,
kind,
is_merge,
chunks,
)?;
return Ok(());
}
let single = blocks
.into_iter()
.next()
.ok_or_else(|| AgentError::Internal("cannot split empty summary block".into()))?;
let (left, right) = Self::split_text_in_half(&single)?;
self.collect_fitting_chunks(
vec![left],
input_budget,
max_summary_tokens,
kind,
is_merge,
chunks,
)?;
self.collect_fitting_chunks(
vec![right],
input_budget,
max_summary_tokens,
kind,
is_merge,
chunks,
)?;
Ok(())
}
fn build_prompt(
&self,
kind: SummaryKind,
is_merge: bool,
body: &str,
max_summary_tokens: usize,
) -> String {
match (kind, is_merge) {
(SummaryKind::Conversation, false) => format!(
"Summarise the following conversation concisely, preserving all key facts, \
decisions, and any pending or in-progress work. \
The summary MUST NOT exceed {} tokens. \
Use this format:\n\n\
**Summary:** <one-paragraph overview>\n\
**Key decisions:** <bullet list or 'none'>\n\
**Open items:** <bullet list or 'none'>\n\n\
Conversation:\n\n{}",
max_summary_tokens, body
),
(SummaryKind::Conversation, true) => format!(
"Merge the following partial conversation summaries into a single concise summary. \
Deduplicate overlap, preserve chronology, and keep all concrete decisions, \
status updates, and unresolved work. The summary MUST NOT exceed {} tokens. \
Use this format:\n\n\
**Summary:** <one-paragraph overview>\n\
**Key decisions:** <bullet list or 'none'>\n\
**Open items:** <bullet list or 'none'>\n\n\
Partial summaries:\n\n{}",
max_summary_tokens, body
),
(SummaryKind::RoomIncrement, false) => format!(
"Create an incremental room summary from the new messages below. \
Deduplicate repeated messages, clean noise, keep chronological order, and preserve \
decisions, facts, assignments/owners, unresolved questions, and concrete next steps. \
The result MUST NOT exceed {} tokens.\n\n\
Format:\n\
**Summary:** <compact overview>\n\
**Decisions:** <bullets or 'none'>\n\
**Owners:** <bullets with owner -> task or 'none'>\n\
**Open items:** <bullets or 'none'>\n\n\
New messages:\n\n{}",
max_summary_tokens, body
),
(SummaryKind::RoomIncrement, true) => format!(
"Merge the following partial room summaries into one room summary. Deduplicate overlap, \
keep chronology, preserve decisions, facts, assignments/owners, unresolved questions, \
and concrete next steps. The result MUST NOT exceed {} tokens.\n\n\
Format:\n\
**Summary:** <compact overview>\n\
**Decisions:** <bullets or 'none'>\n\
**Owners:** <bullets with owner -> task or 'none'>\n\
**Open items:** <bullets or 'none'>\n\n\
Partial summaries:\n\n{}",
max_summary_tokens, body
),
}
}
fn join_blocks(blocks: &[String], is_merge: bool) -> String {
if is_merge {
blocks
.iter()
.enumerate()
.map(|(index, block)| format!("### Partial Summary {}\n{}", index + 1, block))
.collect::<Vec<_>>()
.join("\n\n")
} else {
blocks.join("\n")
}
}
fn split_text_in_half(text: &str) -> Result<(String, String), AgentError> {
if text.chars().count() < 2 {
return Err(AgentError::Internal(
"single summary block exceeds input budget and cannot be split".into(),
));
}
let midpoint = text.len() / 2;
let mut split_at = text.floor_char_boundary(midpoint);
if split_at == 0 || split_at >= text.len() {
split_at = text.ceil_char_boundary(midpoint);
}
if split_at == 0 || split_at >= text.len() {
return Err(AgentError::Internal(
"failed to split oversized summary block".into(),
));
}
Ok((text[..split_at].to_string(), text[split_at..].to_string()))
}
fn estimate_tokens(&self, text: &str) -> usize {
count_message_text(text, &self.model).unwrap_or_else(|_| (text.len() / 4).max(1))
}
fn safe_model_input_budget(&self) -> usize {
Self::safe_model_input_budget_from_limit(self.model_context_limit)
}
fn final_summary_budget(max_summary_tokens: usize) -> usize {
max_summary_tokens.clamp(
CompactConfig::MIN_SUMMARY_TOKENS,
CompactConfig::MAX_SUMMARY_TOKENS,
)
}
fn round_summary_budget(final_budget: usize, input_budget: usize) -> usize {
final_budget.min((input_budget / 8).max(MIN_ROUND_SUMMARY_TOKENS))
}
fn temperature_for(kind: SummaryKind) -> f32 {
match kind {
SummaryKind::Conversation => 0.3,
SummaryKind::RoomIncrement => 0.2,
}
}
fn safe_model_input_budget_from_limit(model_context_limit: Option<usize>) -> usize {
let context_limit = model_context_limit
.unwrap_or(DEFAULT_MODEL_CONTEXT_LIMIT)
.max(1);
context_limit
.saturating_mul(MODEL_INPUT_RATIO_NUMERATOR)
.saturating_div(MODEL_INPUT_RATIO_DENOMINATOR)
.max(1)
}
fn accumulate_usage(total: &mut TokenUsage, has_usage: &mut bool, usage: Option<TokenUsage>) {
if let Some(usage) = usage {
total.input_tokens += usage.input_tokens;
total.output_tokens += usage.output_tokens;
*has_usage = true;
}
}
}
#[cfg(test)]
mod tests {
use super::super::CompactService;
#[test]
fn room_summary_uses_eighty_five_percent_input_budget() {
assert_eq!(
CompactService::safe_model_input_budget_from_limit(Some(1000)),
850
);
}
#[test]
fn oversized_text_is_split_in_half() {
let (left, right) = CompactService::split_text_in_half("abcdefgh").unwrap();
assert_eq!(format!("{}{}", left, right), "abcdefgh");
assert!(!left.is_empty());
assert!(!right.is_empty());
}
}

209
libs/agent/compact/types.rs
View File

@ -1,209 +0,0 @@
use chrono::{DateTime, Utc};
use models::rooms::{
MessageContentType, MessageSenderType, room_message::Model as RoomMessageModel,
};
use serde::{Deserialize, Serialize};
use serde_json::Value;
use uuid::Uuid;
use crate::tokent::TokenUsage;
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct MessageSummary {
pub id: Uuid,
pub sender_type: MessageSenderType,
pub sender_id: Option<Uuid>,
pub sender_name: String,
pub content: String,
pub content_type: MessageContentType,
/// Tool call ID extracted from message content JSON, if present.
pub tool_call_id: Option<String>,
pub send_at: DateTime<Utc>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CompactSummary {
pub session_id: Uuid,
pub room_id: Uuid,
pub retained: Vec<MessageSummary>,
pub summary: String,
pub compacted_at: DateTime<Utc>,
pub messages_compressed: usize,
/// Token usage for the compaction AI call. `None` if usage data was unavailable.
pub usage: Option<TokenUsage>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct RoomCompactRecord {
pub id: Uuid,
pub room_id: Uuid,
pub from_seq: i64,
pub to_seq: i64,
pub summary: String,
pub message_count: i32,
pub source_message_ids: Vec<Uuid>,
pub created_at: DateTime<Utc>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct RoomCompactContext {
pub room_id: Uuid,
pub cutoff_seq: Option<i64>,
pub summary: Option<String>,
pub retained: Vec<MessageSummary>,
pub estimated_tokens: usize,
pub compacted: bool,
}
#[derive(Debug, Clone, Copy)]
pub enum CompactLevel {
Light,
Aggressive,
}
impl CompactLevel {
pub fn retain_count(&self) -> usize {
match self {
CompactLevel::Light => 5,
CompactLevel::Aggressive => 2,
}
}
/// Auto-select level based on estimated token count and config.
///
/// - `Light` (retain 5): when tokens are moderately over threshold
/// - `Aggressive` (retain 2): when tokens are severely over threshold (2x+)
pub fn auto_select(estimated_tokens: usize, threshold: usize) -> Self {
if threshold == 0 {
return CompactLevel::Light;
}
if estimated_tokens >= threshold * 2 {
CompactLevel::Aggressive
} else {
CompactLevel::Light
}
}
}
/// Configuration for automatic compaction.
#[derive(Debug, Clone, Copy)]
pub struct CompactConfig {
/// Only trigger compaction when estimated token count exceeds this.
/// Set to 0 to disable threshold (always compact when messages > retain_count).
pub token_threshold: usize,
/// If true, auto-select level based on how far over the threshold we are.
/// If false, always use `default_level`.
pub auto_level: bool,
/// Fallback level when `auto_level` is false.
pub default_level: CompactLevel,
/// Maximum tokens the summary may contain (enforced via prompt).
pub max_summary_tokens: usize,
}
impl Default for CompactConfig {
fn default() -> Self {
Self {
token_threshold: 100_000,
auto_level: true,
default_level: CompactLevel::Light,
max_summary_tokens: 4096,
}
}
}
impl CompactConfig {
pub const MIN_SUMMARY_TOKENS: usize = 256;
pub const MAX_SUMMARY_TOKENS: usize = 4096;
/// Build config from project context settings.
pub fn from_project_setting(
context_window_tokens: i32,
compaction_threshold: f32,
compaction_max_summary_ratio: f32,
) -> Self {
let context_window_tokens = context_window_tokens.max(0) as usize;
let threshold = (context_window_tokens as f32 * compaction_threshold.max(0.0)) as usize;
Self {
token_threshold: threshold,
auto_level: true,
default_level: CompactLevel::Light,
max_summary_tokens: Self::summary_token_budget(
context_window_tokens,
compaction_max_summary_ratio,
),
}
}
pub fn summary_token_budget(
context_window_tokens: usize,
compaction_max_summary_ratio: f32,
) -> usize {
let ratio = compaction_max_summary_ratio.max(0.0);
let raw_budget = (context_window_tokens as f32 * ratio) as usize;
if raw_budget == 0 {
Self::MIN_SUMMARY_TOKENS
} else {
raw_budget.clamp(Self::MIN_SUMMARY_TOKENS, Self::MAX_SUMMARY_TOKENS)
}
}
}
/// Result of a threshold check before deciding whether to compact.
#[derive(Debug)]
pub enum ThresholdResult {
/// Token count is below threshold — skip compaction.
Skip { estimated_tokens: usize },
/// Token count exceeds threshold — compact with this level.
Compact {
estimated_tokens: usize,
level: CompactLevel,
},
}
impl From<RoomMessageModel> for MessageSummary {
fn from(m: RoomMessageModel) -> Self {
let sender_type = m.sender_type.clone();
let content = m.content.clone();
Self {
id: m.id,
sender_type: sender_type.clone(),
sender_id: m.sender_id,
sender_name: sender_type.to_string(),
content,
content_type: m.content_type.clone(),
tool_call_id: Self::extract_tool_call_id(&m.content),
send_at: m.send_at,
}
}
}
impl MessageSummary {
fn extract_tool_call_id(content: &str) -> Option<String> {
let content = content.trim();
if let Ok(v) = serde_json::from_str::<Value>(content) {
v.get("tool_call_id")
.and_then(|v| v.as_str())
.map(|s| s.to_string())
} else {
None
}
}
}
#[cfg(test)]
mod tests {
use super::CompactConfig;
#[test]
fn summary_budget_has_minimum_floor() {
assert_eq!(CompactConfig::summary_token_budget(0, 0.0), 256);
assert_eq!(CompactConfig::summary_token_budget(128_000, 0.0), 256);
assert_eq!(CompactConfig::summary_token_budget(1_000, 0.01), 256);
}
#[test]
fn summary_budget_is_capped() {
assert_eq!(CompactConfig::summary_token_budget(128_000, 0.2), 4096);
}
}

63
libs/agent/embed/chunk.rs
View File

@ -1,63 +0,0 @@
/// Maximum characters per chunk for embedding (approximates token limit).
/// text-embedding-3-small: 8192 token limit.
/// For CJK ~1 char/token, for English ~4 chars/token.
/// Conservative limit: 7000 chars to leave room for all languages.
const MAX_CHUNK_CHARS: usize = 7000;
/// Split long text into chunks at paragraph/sentence boundaries.
/// Returns at least one chunk even for empty text.
/// Safe for multi-byte characters (uses char indices, not byte indices).
pub fn chunk_text(text: &str) -> Vec<String> {
if text.is_empty() {
return vec![String::new()];
}
if text.len() <= MAX_CHUNK_CHARS {
return vec![text.to_string()];
}
let char_indices: Vec<usize> = text.char_indices().map(|(i, _)| i).collect();
let total_chars = char_indices.len();
let mut chunks = Vec::new();
let mut start_idx = 0;
while start_idx < total_chars {
let byte_start = char_indices[start_idx];
let end_char_idx = (start_idx + MAX_CHUNK_CHARS).min(total_chars);
let byte_end_candidate = char_indices[end_char_idx - 1]
+ text[char_indices[end_char_idx - 1]..]
.chars()
.next()
.map(|c| c.len_utf8())
.unwrap_or(1);
if end_char_idx >= total_chars {
chunks.push(text[byte_start..].to_string());
break;
}
let search_range = &text[byte_start..byte_end_candidate];
let break_at = search_range
.rfind("\n\n")
.map(|pos| pos + 2)
.or_else(|| search_range.rfind('\n').map(|pos| pos + 1))
.or_else(|| search_range.rfind(". ").map(|pos| pos + 1))
.or_else(|| search_range.rfind("! ").map(|pos| pos + 1))
.or_else(|| search_range.rfind("? ").map(|pos| pos + 1));
if let Some(offset) = break_at {
let byte_end = byte_start + offset;
chunks.push(text[byte_start..byte_end].to_string());
let mut advance = start_idx + 1;
while advance < total_chars && char_indices[advance] < byte_end {
advance += 1;
}
start_idx = advance;
} else {
chunks.push(text[byte_start..byte_end_candidate].to_string());
start_idx = end_char_idx;
}
}
chunks
}

291
libs/agent/embed/client.rs
View File

@ -1,291 +0,0 @@
use rig::client::EmbeddingsClient;
use rig::embeddings::EmbeddingModel;
use rig::providers::openai::Client as OpenAiClient;
use serde::{Deserialize, Serialize};
use crate::embed::qdrant::QdrantClient;
pub struct EmbedClient {
openai: OpenAiClient,
qdrant: QdrantClient,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct EmbedVector {
pub id: String,
pub vector: Vec<f32>,
pub payload: EmbedPayload,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct EmbedPayload {
pub entity_type: String,
pub entity_id: String,
pub text: String,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub extra: Option<serde_json::Value>,
}
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct SearchResult {
pub id: String,
pub score: f32,
pub payload: EmbedPayload,
}
impl EmbedClient {
pub fn new(openai: OpenAiClient, qdrant: QdrantClient) -> Self {
Self { openai, qdrant }
}
pub async fn embed_text(&self, text: &str, model: &str) -> crate::Result<Vec<f32>> {
let model = self.openai.embedding_model(model);
let embeddings = model
.embed_texts(vec![text.to_string()])
.await
.map_err(|e| crate::AgentError::OpenAi(format!("embedding failed: {}", e)))?;
embeddings
.first()
.map(|e| e.vec.iter().map(|v| *v as f32).collect())
.ok_or_else(|| crate::AgentError::OpenAi("no embedding returned".into()))
}
pub async fn embed_batch(&self, texts: &[String], model: &str) -> crate::Result<Vec<Vec<f32>>> {
let model = self.openai.embedding_model(model);
let embeddings = model
.embed_texts(texts.to_vec())
.await
.map_err(|e| crate::AgentError::OpenAi(format!("embedding batch failed: {}", e)))?;
tracing::debug!(
input_count = texts.len(),
returned_count = embeddings.len(),
"embed_batch: API returned"
);
let mut result = vec![Vec::new(); texts.len()];
for (idx, embedding) in embeddings.into_iter().enumerate() {
if idx < result.len() {
result[idx] = embedding.vec.iter().map(|v| *v as f32).collect();
continue;
}
tracing::warn!(
idx,
"embed_batch: provider returned more embeddings than requested"
);
break;
}
// Check for empty results
let empty_count = result.iter().filter(|v| v.is_empty()).count();
if empty_count > 0 {
tracing::warn!(
empty_count = empty_count,
total = texts.len(),
"embed_batch: some embeddings returned empty vectors"
);
}
Ok(result)
}
pub async fn upsert(&self, points: Vec<EmbedVector>) -> crate::Result<()> {
self.qdrant.upsert_points(points).await
}
/// Upsert points into a named collection (bypasses entity_type routing).
pub async fn upsert_to_collection(
&self,
collection_name: &str,
points: Vec<EmbedVector>,
) -> crate::Result<()> {
self.qdrant
.upsert_to_collection(collection_name, points)
.await
}
pub async fn search(
&self,
query: &str,
entity_type: &str,
model: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
let vector = self.embed_text(query, model).await?;
self.qdrant.search(&vector, entity_type, limit).await
}
pub async fn search_with_filter(
&self,
query: &str,
entity_type: &str,
model: &str,
limit: usize,
filter: qdrant_client::qdrant::Filter,
) -> crate::Result<Vec<SearchResult>> {
let vector = self.embed_text(query, model).await?;
self.qdrant
.search_with_filter(&vector, entity_type, limit, filter)
.await
}
pub async fn delete_by_entity_id(
&self,
entity_type: &str,
entity_id: &str,
) -> crate::Result<()> {
self.qdrant.delete_by_filter(entity_type, entity_id).await
}
pub async fn ensure_collection(&self, entity_type: &str, dimensions: u64) -> crate::Result<()> {
self.qdrant.ensure_collection(entity_type, dimensions).await
}
pub async fn ensure_skill_collection(&self, dimensions: u64) -> crate::Result<()> {
self.qdrant.ensure_skill_collection(dimensions).await
}
/// Ensure a room-specific memory collection exists.
pub async fn ensure_room_memory_collection(
&self,
project_name: &str,
room_id: &str,
dimensions: u64,
) -> crate::Result<()> {
self.qdrant
.ensure_room_memory_collection(project_name, room_id, dimensions)
.await
}
/// Embed and store a conversation memory (message) in Qdrant.
/// Uses per-room collection: `room:{project_name}:{room_id}`.
pub async fn embed_memory(
&self,
id: &str,
text: &str,
project_name: &str,
room_id: &str,
user_id: Option<&str>,
model: &str,
) -> crate::Result<()> {
// Compute embedding first to know dimensions
let vector = self.embed_text(text, model).await?;
let collection =
crate::embed::qdrant::QdrantClient::room_memory_collection_name(project_name, room_id);
// Auto-create the room collection with correct dimensions
self.qdrant
.ensure_room_memory_collection(project_name, room_id, vector.len() as u64)
.await?;
let point = EmbedVector {
id: id.to_string(),
vector,
payload: EmbedPayload {
entity_type: "memory".to_string(),
entity_id: room_id.to_string(),
text: text.to_string(),
extra: serde_json::json!({ "user_id": user_id }).into(),
},
};
self.qdrant
.upsert_to_collection(&collection, vec![point])
.await
}
/// Search memory embeddings by semantic similarity within a room.
/// Searches the per-room collection directly — no post-filtering needed.
pub async fn search_memories(
&self,
query: &str,
model: &str,
project_name: &str,
room_id: &str,
limit: usize,
dimensions: u64,
) -> crate::Result<Vec<SearchResult>> {
let vector = self.embed_text(query, model).await?;
let collection =
crate::embed::qdrant::QdrantClient::room_memory_collection_name(project_name, room_id);
// Ensure collection exists (will be no-op if already created)
self.qdrant
.ensure_room_memory_collection(project_name, room_id, dimensions)
.await?;
self.qdrant
.search_collection(&collection, &vector, limit)
.await
}
pub async fn search_memories_after_seq(
&self,
query: &str,
model: &str,
project_name: &str,
room_id: &str,
limit: usize,
dimensions: u64,
after_seq: Option<i64>,
) -> crate::Result<Vec<SearchResult>> {
let fetch_limit = if after_seq.is_some() {
limit.saturating_mul(4).max(limit)
} else {
limit
};
let mut results = self
.search_memories(query, model, project_name, room_id, fetch_limit, dimensions)
.await?;
if let Some(cutoff) = after_seq {
results.retain(|r| {
r.payload
.extra
.as_ref()
.and_then(|v| v.get("seq"))
.and_then(|v| v.as_i64())
.map(|seq| seq > cutoff)
.unwrap_or(false)
});
}
results.truncate(limit);
Ok(results)
}
/// Embed and store a skill in Qdrant.
pub async fn embed_skill(
&self,
id: &str,
name: &str,
description: &str,
content: &str,
project_uuid: &str,
model: &str,
) -> crate::Result<()> {
let text = format!("{}: {} {}", name, description, content);
let vector = self.embed_text(&text, model).await?;
let point = EmbedVector {
id: id.to_string(),
vector,
payload: EmbedPayload {
entity_type: "skill".to_string(),
entity_id: project_uuid.to_string(),
text,
extra: serde_json::json!({ "name": name, "description": description }).into(),
},
};
self.qdrant.upsert_points(vec![point]).await
}
/// Search skill embeddings by semantic similarity within a project.
pub async fn search_skills(
&self,
query: &str,
model: &str,
project_uuid: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
let vector = self.embed_text(query, model).await?;
let mut results = self.qdrant.search_skill(&vector, limit + 1).await?;
results.retain(|r| r.payload.entity_id == project_uuid);
results.truncate(limit);
Ok(results)
}
}

24
libs/agent/embed/embeddable.rs
View File

@ -1,24 +0,0 @@
use async_trait::async_trait;
/// Trait for entities that can be embedded as vectors into Qdrant.
#[async_trait]
pub trait Embeddable {
fn entity_type(&self) -> &'static str;
fn to_text(&self) -> String;
fn entity_id(&self) -> String;
}
/// Input struct for batch memory embedding into per-room Qdrant collections.
#[derive(Debug, Clone)]
pub struct EmbedMemoryInput {
pub message_id: String,
pub seq: i64,
pub content: String,
pub project_name: String,
pub room_id: String,
pub user_id: Option<String>,
pub sender_type: String,
}
/// Input struct for batch tag embedding.
pub use models::TagEmbedInput;

369
libs/agent/embed/entity_embed.rs
View File

@ -1,369 +0,0 @@
use std::collections::HashMap;
use super::chunk::chunk_text;
use super::client::{EmbedPayload, EmbedVector};
use super::embeddable::{EmbedMemoryInput, Embeddable};
/// Embedding and upsert operations for entity vectors in Qdrant.
impl super::EmbedService {
pub async fn embed_issue(
&self,
id: &str,
title: &str,
body: Option<&str>,
) -> crate::Result<()> {
let text = match body {
Some(b) if !b.is_empty() => format!("{}\n\n{}", title, b),
_ => title.to_string(),
};
tracing::debug!(issue_id = %id, text_len = text.len(), "embed_issue: calling embedding API");
let vector = self.client.embed_text(&text, &self.model_name).await?;
tracing::debug!(issue_id = %id, vec_dim = vector.len(), "embed_issue: embedding done");
let point = EmbedVector {
id: id.to_string(),
vector,
payload: EmbedPayload {
entity_type: "issue".to_string(),
entity_id: id.to_string(),
text,
extra: None,
},
};
self.client.upsert(vec![point]).await?;
tracing::info!(issue_id = %id, "embed_issue: upsert complete");
Ok(())
}
pub async fn embed_repo(
&self,
id: &str,
name: &str,
description: Option<&str>,
) -> crate::Result<()> {
let text = match description {
Some(d) if !d.is_empty() => format!("{}: {}", name, d),
_ => name.to_string(),
};
tracing::debug!(repo_id = %id, text_len = text.len(), "embed_repo: calling embedding API");
let vector = self.client.embed_text(&text, &self.model_name).await?;
tracing::debug!(repo_id = %id, vec_dim = vector.len(), "embed_repo: embedding done");
let point = EmbedVector {
id: id.to_string(),
vector,
payload: EmbedPayload {
entity_type: "repo".to_string(),
entity_id: id.to_string(),
text,
extra: None,
},
};
self.client.upsert(vec![point]).await?;
tracing::info!(repo_id = %id, "embed_repo: upsert complete");
Ok(())
}
pub async fn embed_issues<T: Embeddable + Send + Sync>(
&self,
items: Vec<T>,
) -> crate::Result<()> {
if items.is_empty() {
return Ok(());
}
let texts: Vec<String> = items.iter().map(|i| i.to_text()).collect();
tracing::debug!(count = texts.len(), "embed_issues: calling embed_batch");
let embeddings = self.client.embed_batch(&texts, &self.model_name).await?;
tracing::debug!(count = embeddings.len(), "embed_issues: batch done");
let points: Vec<EmbedVector> = items
.into_iter()
.zip(embeddings.into_iter())
.map(|(item, vector)| EmbedVector {
id: item.entity_id(),
vector,
payload: EmbedPayload {
entity_type: item.entity_type().to_string(),
entity_id: item.entity_id(),
text: item.to_text(),
extra: None,
},
})
.collect();
let count = points.len();
self.client.upsert(points).await?;
tracing::info!(count = count, "embed_issues: upsert complete");
Ok(())
}
pub async fn embed_skill(
&self,
skill_id: i64,
name: &str,
description: Option<&str>,
content: &str,
project_uuid: &str,
) -> crate::Result<()> {
let desc = description.unwrap_or_default();
let id = skill_id.to_string();
tracing::debug!(skill_id = %skill_id, name = %name, content_len = content.len(), "embed_skill: starting");
let texts = chunk_text(content);
tracing::debug!(skill_id = %skill_id, chunks = texts.len(), "embed_skill: chunked");
if texts.len() == 1 {
self.client
.embed_skill(&id, name, desc, content, project_uuid, &self.model_name)
.await?;
} else {
let full_texts: Vec<String> = texts
.iter()
.map(|t| format!("{}: {} {}", name, desc, t))
.collect();
tracing::debug!(skill_id = %skill_id, "embed_skill: calling embed_batch");
let embeddings = self
.client
.embed_batch(&full_texts, &self.model_name)
.await?;
let points: Vec<EmbedVector> = embeddings
.into_iter()
.enumerate()
.map(|(i, vector)| EmbedVector {
id: format!("{}:chunk:{}", id, i),
vector,
payload: EmbedPayload {
entity_type: "skill".to_string(),
entity_id: project_uuid.to_string(),
text: texts[i].clone(),
extra: serde_json::json!({
"name": name,
"description": desc,
"chunk_index": i,
"total_chunks": texts.len(),
})
.into(),
},
})
.collect();
self.client.upsert(points).await?;
}
tracing::info!(skill_id = %skill_id, chunks = texts.len(), "embed_skill: complete");
Ok(())
}
pub async fn embed_issue_chunked(
&self,
id: &str,
title: &str,
body: Option<&str>,
) -> crate::Result<()> {
let text = match body {
Some(b) if !b.is_empty() => format!("{}\n\n{}", title, b),
_ => title.to_string(),
};
let chunks = chunk_text(&text);
if chunks.len() == 1 {
return self.embed_issue(id, title, body).await;
}
let embeddings = self.client.embed_batch(&chunks, &self.model_name).await?;
let points: Vec<EmbedVector> = embeddings
.into_iter()
.enumerate()
.map(|(i, vector)| EmbedVector {
id: format!("{}:chunk:{}", id, i),
vector,
payload: EmbedPayload {
entity_type: "issue".to_string(),
entity_id: id.to_string(),
text: chunks[i].clone(),
extra: serde_json::json!({
"chunk_index": i,
"total_chunks": chunks.len(),
})
.into(),
},
})
.collect();
self.client.upsert(points).await
}
pub async fn embed_memories_batch(&self, messages: Vec<EmbedMemoryInput>) -> crate::Result<()> {
if messages.is_empty() {
return Ok(());
}
let mut by_room: HashMap<String, Vec<(EmbedMemoryInput, Vec<String>)>> = HashMap::new();
for msg in messages {
let chunks = chunk_text(&msg.content);
if chunks.is_empty() || chunks.iter().all(|c| c.trim().is_empty()) {
continue;
}
let collection = super::qdrant::QdrantClient::room_memory_collection_name(
&msg.project_name,
&msg.room_id,
);
by_room.entry(collection).or_default().push((msg, chunks));
}
for (collection, entries) in &by_room {
let all_texts: Vec<String> = entries
.iter()
.flat_map(|(_, chunks)| chunks.iter().cloned())
.collect();
if all_texts.is_empty() {
continue;
}
let embeddings = self
.client
.embed_batch(&all_texts, &self.model_name)
.await?;
if let Some((first, _)) = entries.first() {
let _ = self
.client
.ensure_room_memory_collection(
&first.project_name,
&first.room_id,
self.dimensions,
)
.await;
}
let mut points = Vec::new();
let mut embed_idx = 0;
for (msg, chunks) in entries {
for (chunk_i, chunk) in chunks.iter().enumerate() {
if embed_idx >= embeddings.len() {
break;
}
let point_id = if chunks.len() == 1 {
msg.message_id.clone()
} else {
format!("{}:chunk:{}", msg.message_id, chunk_i)
};
points.push(EmbedVector {
id: point_id,
vector: embeddings[embed_idx].clone(),
payload: EmbedPayload {
entity_type: "memory".to_string(),
entity_id: msg.room_id.clone(),
text: chunk.clone(),
extra: serde_json::json!({
"message_id": msg.message_id,
"seq": msg.seq,
"user_id": msg.user_id,
"sender_type": msg.sender_type,
"chunk_index": if chunks.len() > 1 {
Some(chunk_i)
} else {
None
},
"total_chunks": if chunks.len() > 1 {
Some(chunks.len())
} else {
None
},
})
.into(),
},
});
embed_idx += 1;
}
}
if let Err(e) = self.client.upsert_to_collection(collection, points).await {
tracing::warn!(collection = %collection, error = %e, "batch memory embed failed");
}
}
Ok(())
}
pub async fn embed_tags_batch(
&self,
tags: Vec<super::embeddable::TagEmbedInput>,
) -> crate::Result<()> {
if tags.is_empty() {
return Ok(());
}
let texts: Vec<String> = tags
.iter()
.map(|t| {
if let Some(ref desc) = t.description {
if !desc.is_empty() {
format!("{}: {}", t.name, desc)
} else {
t.name.clone()
}
} else {
t.name.clone()
}
})
.collect();
let embeddings = self.client.embed_batch(&texts, &self.model_name).await?;
let points: Vec<EmbedVector> = tags
.into_iter()
.zip(embeddings.into_iter())
.map(|(tag, vector)| {
let point_id = format!("{}:{}", tag.repo_id, tag.name);
EmbedVector {
id: point_id,
vector,
payload: EmbedPayload {
entity_type: "repo_tag".to_string(),
entity_id: tag.project_id.clone(),
text: tag.name.clone(),
extra: serde_json::json!({
"repo_id": tag.repo_id,
"repo_name": tag.repo_name,
"tag_name": tag.name,
"description": tag.description,
})
.into(),
},
}
})
.collect();
self.client.upsert(points).await
}
pub async fn embed_memory(
&self,
message_id: &str,
text: &str,
project_name: &str,
room_id: &str,
user_id: Option<&str>,
) -> crate::Result<()> {
self.client
.embed_memory(
message_id,
text,
project_name,
room_id,
user_id,
&self.model_name,
)
.await
}
}

90
libs/agent/embed/mod.rs
View File

@ -1,90 +0,0 @@
pub mod chunk;
pub mod client;
pub mod embeddable;
pub mod entity_embed;
pub mod qdrant;
pub mod search;
pub use client::{EmbedClient, EmbedPayload, EmbedVector, SearchResult};
pub use embeddable::{EmbedMemoryInput, Embeddable, TagEmbedInput};
pub use qdrant::QdrantClient;
use std::sync::Arc;
#[derive(Clone)]
pub struct EmbedService {
client: Arc<EmbedClient>,
db: sea_orm::DatabaseConnection,
model_name: String,
dimensions: u64,
}
impl EmbedService {
pub fn new(
client: EmbedClient,
db: sea_orm::DatabaseConnection,
model_name: String,
dimensions: u64,
) -> Self {
Self {
client: Arc::new(client),
db,
model_name,
dimensions,
}
}
pub async fn ensure_collections(&self) -> crate::Result<()> {
self.client
.ensure_collection("issue", self.dimensions)
.await?;
self.client
.ensure_collection("repo", self.dimensions)
.await?;
self.client.ensure_skill_collection(self.dimensions).await?;
self.client
.ensure_collection("repo_tag", self.dimensions)
.await?;
Ok(())
}
pub fn db(&self) -> &sea_orm::DatabaseConnection {
&self.db
}
pub fn client(&self) -> &Arc<EmbedClient> {
&self.client
}
pub fn model_name(&self) -> &str {
&self.model_name
}
pub fn dimensions(&self) -> u64 {
self.dimensions
}
}
pub async fn new_embed_client(config: &config::AppConfig) -> crate::Result<EmbedClient> {
let base_url = config
.get_embed_model_base_url()
.map_err(|e| crate::AgentError::Internal(e.to_string()))?;
let api_key = config
.get_embed_model_api_key()
.map_err(|e| crate::AgentError::Internal(e.to_string()))?;
let qdrant_url = config
.get_qdrant_url()
.map_err(|e| crate::AgentError::Internal(e.to_string()))?;
let qdrant_api_key = config.get_qdrant_api_key();
let openai = rig::providers::openai::Client::builder()
.api_key(&api_key)
.base_url(&base_url)
.build()
.map_err(|e| {
crate::AgentError::Internal(format!("failed to build rig openai client: {}", e))
})?;
let qdrant = QdrantClient::new(&qdrant_url, qdrant_api_key.as_deref()).await?;
Ok(EmbedClient::new(openai, qdrant))
}

373
libs/agent/embed/qdrant.rs
View File

@ -1,373 +0,0 @@
use qdrant_client::Qdrant;
use qdrant_client::qdrant::{
Condition, CreateCollectionBuilder, DeletePointsBuilder, Distance, FieldCondition, Filter,
Match, PointStruct, SearchPointsBuilder, UpsertPointsBuilder, VectorParamsBuilder, Vectors,
condition::ConditionOneOf, r#match::MatchValue, point_id::PointIdOptions, value,
};
use std::collections::HashMap;
use std::sync::Arc;
use super::client::{EmbedPayload, SearchResult};
use crate::embed::client::EmbedVector;
pub struct QdrantClient {
inner: Arc<Qdrant>,
}
impl Clone for QdrantClient {
fn clone(&self) -> Self {
Self {
inner: self.inner.clone(),
}
}
}
impl QdrantClient {
pub async fn new(url: &str, api_key: Option<&str>) -> crate::Result<Self> {
let mut builder = Qdrant::from_url(url);
if let Some(key) = api_key {
builder = builder.api_key(key);
}
let inner = builder
.build()
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
Ok(Self {
inner: Arc::new(inner),
})
}
fn collection_name(entity_type: &str) -> String {
format!("embed_{}", entity_type)
}
/// Generate the collection name for a room's memory vectors.
pub fn room_memory_collection_name(project_name: &str, room_id: &str) -> String {
let _ = project_name;
format!("room_memory_{}", room_id.replace('-', "_"))
}
pub async fn ensure_collection(&self, entity_type: &str, dimensions: u64) -> crate::Result<()> {
let name = Self::collection_name(entity_type);
self.ensure_collection_named(&name, dimensions).await
}
async fn ensure_collection_named(&self, name: &str, dimensions: u64) -> crate::Result<()> {
let exists = self
.inner
.collection_exists(name)
.await
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
if exists {
return Ok(());
}
let create_collection = CreateCollectionBuilder::new(name)
.vectors_config(VectorParamsBuilder::new(dimensions, Distance::Cosine))
.build();
self.inner
.create_collection(create_collection)
.await
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
Ok(())
}
/// Ensure a room-specific memory collection exists.
pub async fn ensure_room_memory_collection(
&self,
project_name: &str,
room_id: &str,
dimensions: u64,
) -> crate::Result<()> {
let name = Self::room_memory_collection_name(project_name, room_id);
self.ensure_collection_named(&name, dimensions).await
}
pub async fn upsert_points(&self, points: Vec<EmbedVector>) -> crate::Result<()> {
if points.is_empty() {
return Ok(());
}
// Reject empty vectors — they cause Qdrant to reject the entire batch
let empty_vectors = points.iter().filter(|p| p.vector.is_empty()).count();
if empty_vectors > 0 {
tracing::error!(
empty_count = empty_vectors,
total = points.len(),
"upsert_points: REJECTING points with empty vectors"
);
return Err(crate::AgentError::Qdrant(format!(
"refusing to upsert {} points with empty vectors",
empty_vectors
)));
}
let collection_name = Self::collection_name(&points[0].payload.entity_type);
self.upsert_to_collection(&collection_name, points).await
}
/// Upsert points into a specific collection by name.
pub async fn upsert_to_collection(
&self,
collection_name: &str,
points: Vec<EmbedVector>,
) -> crate::Result<()> {
if points.is_empty() {
return Ok(());
}
let qdrant_points: Vec<PointStruct> = points
.into_iter()
.map(|p| {
let mut payload: HashMap<String, qdrant_client::qdrant::Value> = HashMap::new();
payload.insert("entity_type".to_string(), p.payload.entity_type.into());
payload.insert("entity_id".to_string(), p.payload.entity_id.into());
payload.insert("text".to_string(), p.payload.text.into());
if let Some(extra) = p.payload.extra {
let extra_str = serde_json::to_string(&extra).unwrap_or_default();
payload.insert(
"extra".to_string(),
qdrant_client::qdrant::Value {
kind: Some(qdrant_client::qdrant::value::Kind::StringValue(extra_str)),
},
);
}
PointStruct::new(p.id, Vectors::from(p.vector), payload)
})
.collect();
let upsert = UpsertPointsBuilder::new(collection_name, qdrant_points).build();
self.inner
.upsert_points(upsert)
.await
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
Ok(())
}
fn extract_string(value: &qdrant_client::qdrant::Value) -> String {
match &value.kind {
Some(value::Kind::StringValue(s)) => s.clone(),
_ => String::new(),
}
}
pub async fn search(
&self,
vector: &[f32],
entity_type: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
let collection_name = Self::collection_name(entity_type);
self.search_collection(&collection_name, vector, limit)
.await
}
/// Search a specific collection by name.
pub async fn search_collection(
&self,
collection_name: &str,
vector: &[f32],
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
let search_req = SearchPointsBuilder::new(collection_name, vector.to_vec(), limit as u64)
.with_payload(true)
.build();
let results = self
.inner
.search_points(search_req)
.await
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
Ok(results
.result
.into_iter()
.filter_map(|p| {
let entity_type = p
.payload
.get(&"entity_type".to_string())
.map(Self::extract_string)
.unwrap_or_default();
let entity_id = p
.payload
.get(&"entity_id".to_string())
.map(Self::extract_string)
.unwrap_or_default();
let text = p
.payload
.get(&"text".to_string())
.map(Self::extract_string)
.unwrap_or_default();
let extra = p.payload.get(&"extra".to_string()).and_then(|v| {
let s = Self::extract_string(v);
if s.is_empty() {
None
} else {
serde_json::from_str::<serde_json::Value>(&s).ok()
}
});
let id =
p.id.and_then(|id| id.point_id_options)
.map(|opts| match opts {
PointIdOptions::Uuid(s) => s,
PointIdOptions::Num(n) => n.to_string(),
})
.unwrap_or_default();
Some(SearchResult {
id,
score: p.score,
payload: EmbedPayload {
entity_type,
entity_id,
text,
extra,
},
})
})
.collect())
}
pub async fn search_with_filter(
&self,
vector: &[f32],
entity_type: &str,
limit: usize,
filter: Filter,
) -> crate::Result<Vec<SearchResult>> {
let collection_name = Self::collection_name(entity_type);
let search = SearchPointsBuilder::new(collection_name, vector.to_vec(), limit as u64)
.with_payload(true)
.filter(filter)
.build();
let results = self
.inner
.search_points(search)
.await
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
Ok(results
.result
.into_iter()
.filter_map(|p| {
let entity_type = p
.payload
.get(&"entity_type".to_string())
.map(Self::extract_string)
.unwrap_or_default();
let entity_id = p
.payload
.get(&"entity_id".to_string())
.map(Self::extract_string)
.unwrap_or_default();
let text = p
.payload
.get(&"text".to_string())
.map(Self::extract_string)
.unwrap_or_default();
let extra = p.payload.get(&"extra".to_string()).and_then(|v| {
let s = Self::extract_string(v);
if s.is_empty() {
None
} else {
serde_json::from_str::<serde_json::Value>(&s).ok()
}
});
let id =
p.id.and_then(|id| id.point_id_options)
.map(|opts| match opts {
PointIdOptions::Uuid(s) => s,
PointIdOptions::Num(n) => n.to_string(),
})
.unwrap_or_default();
Some(SearchResult {
id,
score: p.score,
payload: EmbedPayload {
entity_type,
entity_id,
text,
extra,
},
})
})
.collect())
}
pub async fn delete_by_filter(&self, entity_type: &str, entity_id: &str) -> crate::Result<()> {
let collection_name = Self::collection_name(entity_type);
let filter = Filter {
must: vec![Condition {
condition_one_of: Some(ConditionOneOf::Field(FieldCondition {
key: "entity_id".to_string(),
r#match: Some(Match {
match_value: Some(MatchValue::Keyword(entity_id.to_string())),
}),
..Default::default()
})),
}],
..Default::default()
};
let delete = DeletePointsBuilder::new(collection_name)
.points(filter)
.build();
self.inner
.delete_points(delete)
.await
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
Ok(())
}
pub async fn delete_collection(&self, entity_type: &str) -> crate::Result<()> {
let name = Self::collection_name(entity_type);
self.inner
.delete_collection(name)
.await
.map_err(|e| crate::AgentError::Qdrant(e.to_string()))?;
Ok(())
}
pub async fn ensure_memory_collection(&self, dimensions: u64) -> crate::Result<()> {
self.ensure_collection("memory", dimensions).await
}
pub async fn ensure_skill_collection(&self, dimensions: u64) -> crate::Result<()> {
self.ensure_collection("skill", dimensions).await
}
pub async fn search_memory(
&self,
vector: &[f32],
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
self.search(vector, "memory", limit).await
}
pub async fn search_skill(
&self,
vector: &[f32],
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
self.search(vector, "skill", limit).await
}
}

107
libs/agent/embed/search.rs
View File

@ -1,107 +0,0 @@
use qdrant_client::qdrant::Filter;
use super::client::SearchResult;
/// Vector search operations for Qdrant-backed entity retrieval.
impl super::EmbedService {
pub async fn search_issues(
&self,
query: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
self.client
.search(query, "issue", &self.model_name, limit)
.await
}
pub async fn search_repos(
&self,
query: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
self.client
.search(query, "repo", &self.model_name, limit)
.await
}
pub async fn search_issues_filtered(
&self,
query: &str,
limit: usize,
filter: Filter,
) -> crate::Result<Vec<SearchResult>> {
self.client
.search_with_filter(query, "issue", &self.model_name, limit, filter)
.await
}
/// Search repo tags by semantic similarity within a project.
/// Filters by project_id (stored in entity_id) for project isolation.
pub async fn search_tags(
&self,
query: &str,
project_id: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
let mut results = self
.client
.search(query, "repo_tag", &self.model_name, limit + 1)
.await?;
results.retain(|r| r.payload.entity_id == project_id);
results.truncate(limit);
Ok(results)
}
/// Search skills by semantic similarity within a project.
pub async fn search_skills(
&self,
query: &str,
project_uuid: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
self.client
.search_skills(query, &self.model_name, project_uuid, limit)
.await
}
/// Search past conversation messages by semantic similarity within a room.
pub async fn search_memories(
&self,
query: &str,
project_name: &str,
room_id: &str,
limit: usize,
) -> crate::Result<Vec<SearchResult>> {
self.client
.search_memories(
query,
&self.model_name,
project_name,
room_id,
limit,
self.dimensions,
)
.await
}
pub async fn search_memories_after_seq(
&self,
query: &str,
project_name: &str,
room_id: &str,
limit: usize,
after_seq: Option<i64>,
) -> crate::Result<Vec<SearchResult>> {
self.client
.search_memories_after_seq(
query,
&self.model_name,
project_name,
room_id,
limit,
self.dimensions,
after_seq,
)
.await
}
}

63
libs/agent/error.rs
View File

@ -1,63 +0,0 @@
use thiserror::Error;
#[derive(Error, Debug)]
pub enum AgentError {
#[error("openai error: {0}")]
OpenAi(String),
#[error("qdrant error: {0}")]
Qdrant(String),
#[error("internal error: {0}")]
Internal(String),
#[error("not found: {0}")]
NotFound(String),
/// The task exceeded its timeout limit.
#[error("task {task_id} timed out after {seconds}s")]
Timeout { task_id: i64, seconds: u64 },
/// The agent has been rate-limited; retry after the indicated delay.
#[error("rate limited, retry after {retry_after_secs}s")]
RateLimited { retry_after_secs: u64 },
/// A transient error that can be retried.
#[error("retryable error (attempt {attempt}): {message}")]
Retryable { attempt: u32, message: String },
/// The requested tool is not registered in the tool registry.
#[error("tool not found: {tool}")]
ToolNotFound { tool: String },
/// A tool execution failed.
#[error("tool '{tool}' execution failed: {cause}")]
ToolExecutionFailed { tool: String, cause: String },
/// The request contains invalid input.
#[error("invalid input in '{field}': {reason}")]
InvalidInput { field: String, reason: String },
}
pub type Result<T> = std::result::Result<T, AgentError>;
impl From<qdrant_client::QdrantError> for AgentError {
fn from(e: qdrant_client::QdrantError) -> Self {
AgentError::Qdrant(e.to_string())
}
}
impl From<sea_orm::DbErr> for AgentError {
fn from(e: sea_orm::DbErr) -> Self {
AgentError::Internal(e.to_string())
}
}
impl From<crate::tool::ToolError> for AgentError {
fn from(e: crate::tool::ToolError) -> Self {
AgentError::ToolExecutionFailed {
tool: String::new(),
cause: e.to_string(),
}
}
}

60
libs/agent/lib.rs
View File

@ -1,60 +0,0 @@
pub mod agent;
pub mod billing;
pub mod chat;
pub mod client;
pub mod compact;
pub mod embed;
pub mod error;
pub mod model;
pub mod orao;
pub mod perception;
pub mod react;
pub mod skills;
pub mod sync;
pub mod task;
pub mod tokent;
pub mod tool;
pub use billing::{
BillingRecord, BillingResult, check_balance, check_user_balance, initialize_project_billing,
initialize_user_billing, persist_billing_error, record_ai_usage, record_user_ai_usage,
};
pub use chat::{
AgentExecutionProfile, AgentRole, AiContextSenderType, AiRequest, AiStreamChunk, ChatService,
Mention, RoomMessageContext, StreamCallback,
};
pub use client::types::ChatRequestMessage;
pub use client::{AiCallResponse, AiClientConfig, call_with_params, call_with_retry};
pub use compact::{
CompactConfig, CompactLevel, CompactService, CompactSummary, MessageSummary,
RoomCompactContext, RoomCompactRecord,
};
pub use embed::{
EmbedClient, EmbedMemoryInput, EmbedService, QdrantClient, SearchResult, TagEmbedInput,
new_embed_client,
};
pub use error::{AgentError, Result};
pub use orao::{
ActionExecutor, ActionResult, ActionType, ActionVerdict, OraoConfig, OraoExecutor,
OraoExecutorBuilder, OraoOutcome, OraoStep, PerceptionSnapshot, PlannedAction, ReasoningOutput,
RoundRecord, SafetyLevel,
};
pub use perception::{PerceptionService, SkillContext, SkillEntry, ToolCallEvent};
pub use react::{
DEFAULT_SYSTEM_PROMPT, PERSONAL_CONTEXT_PROMPT, ROOM_CONTEXT_PROMPT, ReactConfig, ReactStep,
};
pub use skills::{
BuiltInSkill, SKILL_TEMPLATES, all_skill_slugs, all_skills, get_skill, get_skill_by_tool,
is_built_in_skill, match_skill_by_keyword, skills_by_category,
};
pub use sync::list_accessible_models;
pub use task::TaskService;
pub use tokent::{TokenUsage, resolve_usage};
pub use tool::{
ToolCall, ToolCallRecord, ToolCallRecorder, ToolCallResult, ToolContext, ToolDefinition,
ToolError, ToolExecutor, ToolHandler, ToolParam, ToolRegistry, ToolResult, ToolSchema,
};
#[cfg(feature = "rig")]
pub use agent::RigAgentService;
#[cfg(feature = "rig")]
pub use tool::{RecordingTool, RigToolSet, is_retryable_tool_error};

117
libs/agent/model/capability.rs
View File

@ -1,117 +0,0 @@
//! Model capability management — CRUD.
use chrono::Utc;
use db::database::AppDatabase;
use models::agents::CapabilityType;
use models::agents::model_capability;
use sea_orm::*;
use crate::error::AgentError;
#[derive(Debug, Clone, serde::Deserialize, utoipa::ToSchema)]
pub struct CreateModelCapabilityRequest {
pub model_version_id: i64,
pub capability: String,
#[serde(default)]
pub is_supported: bool,
}
#[derive(Debug, Clone, serde::Deserialize, utoipa::ToSchema)]
pub struct UpdateModelCapabilityRequest {
pub is_supported: Option<bool>,
}
#[derive(Debug, Clone, serde::Serialize, utoipa::ToSchema)]
pub struct ModelCapabilityResponse {
pub id: i64,
pub model_version_id: i64,
pub capability: String,
pub is_supported: bool,
pub created_at: chrono::DateTime<Utc>,
}
impl From<model_capability::Model> for ModelCapabilityResponse {
fn from(mc: model_capability::Model) -> Self {
Self {
id: mc.id,
model_version_id: mc.model_version_id,
capability: mc.capability,
is_supported: mc.is_supported,
created_at: mc.created_at,
}
}
}
pub async fn list_capabilities(
db: &AppDatabase,
model_version_id: i64,
) -> Result<Vec<ModelCapabilityResponse>, AgentError> {
let caps = model_capability::Entity::find()
.filter(model_capability::Column::ModelVersionId.eq(model_version_id))
.order_by_asc(model_capability::Column::Capability)
.all(db)
.await?;
Ok(caps
.into_iter()
.map(ModelCapabilityResponse::from)
.collect())
}
pub async fn get_capability(
db: &AppDatabase,
id: i64,
) -> Result<ModelCapabilityResponse, AgentError> {
let cap = model_capability::Entity::find_by_id(id)
.one(db)
.await?
.ok_or_else(|| AgentError::NotFound(format!("Capability record not found: {}", id)))?;
Ok(ModelCapabilityResponse::from(cap))
}
pub async fn create_capability(
db: &AppDatabase,
request: CreateModelCapabilityRequest,
) -> Result<ModelCapabilityResponse, AgentError> {
let _ = request
.capability
.parse::<CapabilityType>()
.map_err(|_| AgentError::InvalidInput {
field: "capability".into(),
reason: "Invalid capability type".into(),
})?;
let now = Utc::now();
let active = model_capability::ActiveModel {
model_version_id: Set(request.model_version_id),
capability: Set(request.capability),
is_supported: Set(request.is_supported),
created_at: Set(now),
..Default::default()
};
let cap = active.insert(db).await?;
Ok(ModelCapabilityResponse::from(cap))
}
pub async fn update_capability(
db: &AppDatabase,
id: i64,
request: UpdateModelCapabilityRequest,
) -> Result<ModelCapabilityResponse, AgentError> {
let cap = model_capability::Entity::find_by_id(id)
.one(db)
.await?
.ok_or_else(|| AgentError::NotFound(format!("Capability record not found: {}", id)))?;
let mut active: model_capability::ActiveModel = cap.into();
if let Some(is_supported) = request.is_supported {
active.is_supported = Set(is_supported);
}
let cap = active.update(db).await?;
Ok(ModelCapabilityResponse::from(cap))
}
pub async fn delete_capability(db: &AppDatabase, id: i64) -> Result<(), AgentError> {
model_capability::Entity::delete_by_id(id).exec(db).await?;
Ok(())
}

6
libs/agent/model/mod.rs
View File

@ -1,6 +0,0 @@
pub mod capability;
pub mod model_entry;
pub mod parameter_profile;
pub mod pricing;
pub mod provider;
pub mod version;

Some files were not shown because too many files have changed in this diff Show More