fix(billing): use actual tokens in nonstreaming ReAct billing

- ai_react_nonstreaming now passes real input/output tokens to billing
- Was passing hardcoded 0,0 despite destructuring token data
- Also fix unused variable warnings

libs/room/src/service/ai_react_nonstreaming.rs

@@ -33,7 +33,7 @@ pub async fn process_message_ai_react_nonstreaming(
             .await;
 
         match final_answer {
-            Ok((response, input_tokens, output_tokens)) => {
+            Ok((response, _input_tokens, _output_tokens)) => {
                 if let Err(e) = create_and_publish_ai_message(
                     &db,
                     &cache,
@@ -66,15 +66,17 @@ pub async fn process_message_ai_react_nonstreaming(
                     }
 
                     // Record billing (non-fatal)
-                    // TODO: ReAct agent does not track token counts yet; billing with 0/0
+                    if let Err(e) = super::billing::record_ai_usage(
-                    let _ = super::billing::record_ai_usage(
                         &db,
                         project_id,
                         model_id,
-                        0,
+                        _input_tokens,
-                        0,
+                        _output_tokens,
                     )
-                    .await;
+                    .await
+                    {
+                        tracing::warn!(error = %e, "AI billing recording failed");
+                    }
                 }
             }
             Err(e) => {

libs/room/src/service/ai_react_streaming.rs

@@ -155,7 +155,7 @@ pub async fn process_message_ai_react_streaming(
             })
             .await;
 
-        let (final_content, input_tokens, output_tokens, err_msg, should_log) = match result {
+        let (final_content, input_tokens, output_tokens, err_msg, _should_log) = match result {
             Ok((content, input, output)) => (content, input, output, None, false),
             Err(e) => {
                 let msg = format!("[Agent error: {}]", e);