96 lines
3.6 KiB
Python
96 lines
3.6 KiB
Python
|
|
import os, json, httpx, traceback
|
|
from fastapi import FastAPI, HTTPException
|
|
from pydantic import BaseModel
|
|
|
|
AGENT_URL = os.getenv("AGENT_URL", "http://ai-agent:8080")
|
|
OPENAI_MODEL = os.getenv("OPENAI_MODEL", "gpt-4o-mini")
|
|
|
|
def _read_api_key():
|
|
path = os.getenv("OPENAI_API_KEY_FILE", "/run/secrets/openai_api_key")
|
|
if os.path.exists(path):
|
|
return open(path, "r").read().strip()
|
|
return os.getenv("OPENAI_API_KEY", "")
|
|
|
|
SYSTEM_PROMPT = (
|
|
"You are an ops command planner. Convert the user's intent into a STRICT JSON object "
|
|
"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."
|
|
)
|
|
|
|
class ChatIn(BaseModel):
|
|
prompt: str
|
|
|
|
app = FastAPI(title="AI Relay (LLM -> Agent)")
|
|
|
|
_last_debug = {"openai_request": None, "openai_response": None, "agent_request": None, "agent_response": None, "error": None}
|
|
|
|
@app.get("/health")
|
|
def health():
|
|
return {"ok": True, "model": OPENAI_MODEL, "agent_url": AGENT_URL}
|
|
|
|
@app.get("/last-raw")
|
|
def last_raw():
|
|
# Expose last request/response for debugging
|
|
return _last_debug
|
|
|
|
@app.post("/chat")
|
|
async def chat(inp: ChatIn):
|
|
api_key = _read_api_key()
|
|
if not api_key:
|
|
raise HTTPException(500, "Missing OPENAI_API_KEY (env or secret).")
|
|
|
|
url = "https://api.openai.com/v1/chat/completions"
|
|
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
|
|
body = {
|
|
"model": OPENAI_MODEL,
|
|
"response_format": {"type": "json_object"},
|
|
"temperature": 0.1,
|
|
"messages": [
|
|
{"role": "system", "content": SYSTEM_PROMPT},
|
|
{"role": "user", "content": inp.prompt},
|
|
],
|
|
}
|
|
|
|
_last_debug["openai_request"] = {"url": url, "body": body}
|
|
_last_debug["openai_response"] = None
|
|
_last_debug["agent_request"] = None
|
|
_last_debug["agent_response"] = None
|
|
_last_debug["error"] = None
|
|
|
|
try:
|
|
async with httpx.AsyncClient(timeout=30) as client:
|
|
r = await client.post(url, headers=headers, json=body)
|
|
_last_debug["openai_response"] = {"status": r.status_code, "text": r.text[:500]}
|
|
r.raise_for_status()
|
|
data = r.json()
|
|
except httpx.RequestError as e:
|
|
_last_debug["error"] = f"OpenAI network error: {str(e)}"
|
|
raise HTTPException(502, f"OpenAI network error: {e}")
|
|
except httpx.HTTPStatusError as e:
|
|
_last_debug["error"] = f"OpenAI HTTP error: {e.response.text}"
|
|
raise HTTPException(502, f"OpenAI error: {e.response.text}")
|
|
|
|
try:
|
|
content = data["choices"][0]["message"]["content"]
|
|
cmd = json.loads(content)
|
|
except Exception as e:
|
|
_last_debug["error"] = f"Parse error: {str(e)}; raw={str(data)[:300]}"
|
|
raise HTTPException(500, f"Failed to parse model JSON: {e}; raw={str(data)[:300]}")
|
|
|
|
_last_debug["agent_request"] = {"url": f"{AGENT_URL}/command", "json": cmd}
|
|
try:
|
|
async with httpx.AsyncClient(timeout=15) as client:
|
|
r = await client.post(f"{AGENT_URL}/command", json=cmd)
|
|
_last_debug["agent_response"] = {"status": r.status_code, "text": r.text[:500]}
|
|
r.raise_for_status()
|
|
return r.json()
|
|
except httpx.RequestError as e:
|
|
_last_debug["error"] = f"Agent network error: {str(e)}"
|
|
raise HTTPException(502, f"Agent network error: {e}")
|
|
except httpx.HTTPStatusError as e:
|
|
_last_debug["error"] = f"Agent HTTP error: {e.response.text}"
|
|
raise HTTPException(e.response.status_code, f"Agent error: {e.response.text}")
|