ai: use atlasbot internal answers before LLM

backend/atlas_portal/routes/ai.py

@@ -23,6 +23,11 @@ def register(app) -> None:
         if not user_message:
             return jsonify({"error": "message required"}), 400
 
+        atlasbot_reply = _atlasbot_answer(user_message)
+        if atlasbot_reply:
+            elapsed_ms = int((time.time() - started) * 1000)
+            return jsonify({"reply": atlasbot_reply, "latency_ms": elapsed_ms, "source": "atlasbot"})
+
         messages: list[dict[str, str]] = []
         if settings.AI_CHAT_SYSTEM_PROMPT:
             messages.append({"role": "system", "content": settings.AI_CHAT_SYSTEM_PROMPT})
@@ -58,6 +63,25 @@ def register(app) -> None:
     _start_keep_warm()
 
 
+def _atlasbot_answer(message: str) -> str:
+    endpoint = settings.AI_ATLASBOT_ENDPOINT
+    if not endpoint:
+        return ""
+    headers: dict[str, str] = {}
+    if settings.AI_ATLASBOT_TOKEN:
+        headers["X-Internal-Token"] = settings.AI_ATLASBOT_TOKEN
+    try:
+        with httpx.Client(timeout=settings.AI_ATLASBOT_TIMEOUT_SEC) as client:
+            resp = client.post(endpoint, json={"prompt": message}, headers=headers)
+            if resp.status_code != 200:
+                return ""
+            data = resp.json()
+            answer = (data.get("answer") or "").strip()
+            return answer
+    except (httpx.RequestError, ValueError):
+        return ""
+
+
 def _discover_ai_meta() -> dict[str, str]:
     meta = {
         "node": settings.AI_NODE_NAME,
@@ -136,4 +160,3 @@ def _start_keep_warm() -> None:
                 continue
 
     threading.Thread(target=loop, daemon=True, name="ai-keep-warm").start()
-

backend/atlas_portal/settings.py

@@ -26,6 +26,9 @@ AI_CHAT_SYSTEM_PROMPT = os.getenv(
     "You are the Titan Lab assistant for bstein.dev. Be concise and helpful.",
 )
 AI_CHAT_TIMEOUT_SEC = float(os.getenv("AI_CHAT_TIMEOUT_SEC", "20"))
+AI_ATLASBOT_ENDPOINT = os.getenv("AI_ATLASBOT_ENDPOINT", "").strip()
+AI_ATLASBOT_TOKEN = os.getenv("AI_ATLASBOT_TOKEN", "").strip()
+AI_ATLASBOT_TIMEOUT_SEC = float(os.getenv("AI_ATLASBOT_TIMEOUT_SEC", "5"))
 AI_NODE_NAME = os.getenv("AI_CHAT_NODE_NAME") or os.getenv("AI_NODE_NAME") or "ai-cluster"
 AI_GPU_DESC = os.getenv("AI_CHAT_GPU_DESC") or "local GPU (dynamic)"
 AI_PUBLIC_ENDPOINT = os.getenv("AI_PUBLIC_CHAT_ENDPOINT", "https://chat.ai.bstein.dev/api/chat")