first commit

relay/relay.py

@@ -7,7 +7,7 @@ AGENT_URL = os.getenv("AGENT_URL", "http://ai-agent:8080")
 OPENAI_MODEL = os.getenv("OPENAI_MODEL", "gpt-4o-mini")
 
 def _read_api_key():
-    # Prefer file from Docker secret if present
+    # Prefer Docker secret if available
     path = os.getenv("OPENAI_API_KEY_FILE", "/run/secrets/openai_api_key")
     if os.path.exists(path):
         return open(path, "r").read().strip()
@@ -18,7 +18,7 @@ SYSTEM_PROMPT = (
     "with fields: action (scale|restart_service), params (dict). No prose. Examples: "
     '{"action":"scale","params":{"service":"weblabs_php","replicas":3}} '
     'or {"action":"restart_service","params":{"service":"weblabs_php"}}. '
-    "Only produce valid JSON. If unclear, choose the safest no-op."
+    "Only produce valid JSON."
 )
 
 class ChatIn(BaseModel):
@@ -36,14 +36,16 @@ async def chat(inp: ChatIn):
     if not api_key:
         raise HTTPException(500, "Missing OPENAI_API_KEY (env or secret).")
 
-    # Call OpenAI Responses API
-    url = "https://api.openai.com/v1/responses"
+    url = "https://api.openai.com/v1/chat/completions"
     headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
     body = {
         "model": OPENAI_MODEL,
-        "input": f"{SYSTEM_PROMPT}\nUSER: {inp.prompt}",
-        "max_output_tokens": 300,
-        "temperature": 0.1
+        "response_format": {"type": "json_object"},
+        "temperature": 0.1,
+        "messages": [
+            {"role": "system", "content": SYSTEM_PROMPT},
+            {"role": "user", "content": inp.prompt},
+        ],
     }
 
     async with httpx.AsyncClient(timeout=30) as client:
@@ -51,22 +53,12 @@ async def chat(inp: ChatIn):
         if r.status_code >= 400:
             raise HTTPException(502, f"OpenAI error: {r.text}")
         data = r.json()
-        # Responses API returns output in 'output_text' (or tool messages). Try common fields.
-        content = data.get("output_text") or data.get("content") or ""
-        if isinstance(content, list):
-            # Some responses return a list of content parts; take text from first text part
-            for part in content:
-                if part.get("type") in ("output_text", "text") and part.get("text"):
-                    content = part["text"]
-                    break
-        if not isinstance(content, str):
-            content = str(content)
 
-    # Parse JSON from the model output
     try:
+        content = data["choices"][0]["message"]["content"]
         cmd = json.loads(content)
     except Exception as e:
-        raise HTTPException(500, f"Failed to parse model JSON: {e}; content={content[:200]}")
+        raise HTTPException(500, f"Failed to parse model JSON: {e}; raw={str(data)[:300]}")
 
     # Forward to the agent
     async with httpx.AsyncClient(timeout=15) as client: