fix: show budget-limit message on 429 instead of generic error

src/agent_manager/api/static/widget/react/AgentChatApp.tsx

@@ -32,6 +32,8 @@ type MessageEntry = {
   typing?: boolean;
   route?: string[];
   tools?: ToolRecord[];
+  /** When true the message represents a permanent error (e.g. budget exceeded). */
+  error?: boolean;
 };
 
 let nextMessageCounter = 0;
@@ -150,6 +152,9 @@ export function AgentChatApp({ client, config, onAnswer, panelId, titleId }: Age
     setEntries((prev) => prev.map((entry) => (entry.id === id ? update(entry) : entry)));
   }, []);
 
+  const BUDGET_EXCEEDED_MESSAGE =
+    "This conversation has reached its context limit. Please start a new chat to continue.";
+
   const submit = useCallback(
     async (text: string) => {
       const assistantId = nextMessageId("ai");
@@ -167,6 +172,16 @@ export function AgentChatApp({ client, config, onAnswer, panelId, titleId }: Age
         patchEntry(assistantId, (entry) => ({ ...entry, typing: false }));
         onAnswer(finalDetail);
       } catch (error) {
+        if (error instanceof AgentChatHttpError && error.status === 429) {
+          patchEntry(assistantId, () => ({
+            id: assistantId,
+            role: "ai",
+            text: BUDGET_EXCEEDED_MESSAGE,
+            error: true,
+          }));
+          setSending(false);
+          return;
+        }
         try {
           const data = await sendToAgent(text);
           patchEntry(assistantId, () => ({
@@ -181,10 +196,7 @@ export function AgentChatApp({ client, config, onAnswer, panelId, titleId }: Age
           patchEntry(assistantId, () => ({
             id: assistantId,
             role: "ai",
-            text:
-              error instanceof Error && error.message
-                ? `Something went wrong. Please try again.`
-                : "Something went wrong. Please try again.",
+            text: "Something went wrong. Please try again.",
           }));
         }
       } finally {
@@ -256,6 +268,7 @@ export function AgentChatApp({ client, config, onAnswer, panelId, titleId }: Age
                   key={entry.id}
                   from={entry.role === "user" ? "user" : "assistant"}
                   typing={entry.typing}
+                  error={entry.error}
                 >
                   {entry.typing ? (
                     "..."

src/agent_manager/api/static/widget/react/shadcnAiElements.tsx

@@ -44,10 +44,11 @@ export function Message({
   children,
   from,
   typing = false,
-}: PropsWithChildren<{ from: "user" | "assistant"; typing?: boolean }>) {
+  error = false,
+}: PropsWithChildren<{ from: "user" | "assistant"; typing?: boolean; error?: boolean }>) {
   return (
     <div
-      className={cn("msg", from === "user" ? "user" : "ai", typing && "typing")}
+      className={cn("msg", from === "user" ? "user" : "ai", typing && "typing", error && "msg-error")}
       role={typing ? "status" : undefined}
       aria-label={typing ? "Assistant is typing" : undefined}
     >
@@ -56,6 +57,7 @@ export function Message({
   );
 }
 
+
 export function MessageContent({ children }: PropsWithChildren) {
   return <div className="message-content">{children}</div>;
 }

src/agent_manager/api/static/widget/styles/styles.ts

@@ -44,6 +44,8 @@ export function styles(config: AgentChatConfig): string {
     .msg { font-size: 14.5px; line-height: 1.55; word-wrap: break-word; white-space: pre-wrap; }
     .msg.ai { color: #18181b; max-width: 100%; }
     .msg.ai.typing { color: #a1a1aa; letter-spacing: 1px; }
+    .msg.ai.msg-error { color: #991b1b; background: #fff5f5; border: 1px solid #fecaca;
+      border-radius: 10px; padding: 8px 12px; }
     .msg.user { background: #f4f4f5; color: #18181b; border-radius: 18px;
       padding: 10px 14px; margin-left: auto; max-width: 88%; }
     .message-content { min-width: 0; }

tests/agent_manager/conftest.py

@@ -30,7 +30,13 @@ async def run(
         self.prompts.append(message)
         self.histories.append(tuple(history))
         self.contexts.append(context)
-        return RunResult(system_name="stub", visited=["agent"], answer=f"answer:{message}")
+        return RunResult(
+            system_name="stub",
+            visited=["agent"],
+            answer=f"answer:{message}",
+            input_tokens=5,
+            output_tokens=5,
+        )
 
     async def stream(
         self,

tests/agent_manager/test_api.py

@@ -165,3 +165,41 @@ def test_create_accepts_stable_session_and_send_accepts_user(client: TestClient)
     sent = client.post("/conversations/sess-1/messages", json={"message": "hello", "user_id": "u1"})
 
     assert sent.status_code == 200
+
+
+def test_send_returns_429_when_token_budget_exceeded() -> None:
+    """send_message returns 429 when the conversation token budget is exhausted."""
+    app = FastAPI()
+    # max_tokens=1 means the budget is treated as exceeded as soon as any token is recorded.
+    # To trigger the guard we first send a real message (consumes tokens), then send a second one.
+    service = ConversationService(RecordingEngine(), MemoryRepository(), max_tokens=1)
+    app.state.service = service
+    app.include_router(router)
+    client = TestClient(app)
+
+    cid = client.post("/conversations").json()["conversation_id"]
+    # First message succeeds and records token usage in the repository.
+    client.post(f"/conversations/{cid}/messages", json={"message": "first"})
+    # Second message should be rejected because the budget is now exhausted.
+    response = client.post(f"/conversations/{cid}/messages", json={"message": "second"})
+
+    assert response.status_code == 429
+    assert "budget" in response.json()["detail"]
+
+
+def test_stream_returns_429_when_token_budget_exceeded() -> None:
+    """stream_message returns 429 when the conversation token budget is exhausted."""
+    app = FastAPI()
+    service = ConversationService(RecordingEngine(), MemoryRepository(), max_tokens=1)
+    app.state.service = service
+    app.include_router(router)
+    client = TestClient(app)
+
+    cid = client.post("/conversations").json()["conversation_id"]
+    # Consume the budget with a non-streaming send first.
+    client.post(f"/conversations/{cid}/messages", json={"message": "first"})
+    # The streaming endpoint should now return 429.
+    response = client.post(f"/conversations/{cid}/messages/stream", json={"message": "second"})
+
+    assert response.status_code == 429
+    assert "budget" in response.json()["detail"]