77 lines
2.8 KiB
Python
77 lines
2.8 KiB
Python
|
|
import os, json, httpx
|
|
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():
|
|
# Prefer file from Docker secret if present
|
|
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. If unclear, choose the safest no-op."
|
|
)
|
|
|
|
class ChatIn(BaseModel):
|
|
prompt: str
|
|
|
|
app = FastAPI(title="AI Relay (LLM -> Agent)")
|
|
|
|
@app.get("/health")
|
|
def health():
|
|
return {"ok": True}
|
|
|
|
@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).")
|
|
|
|
# Call OpenAI Responses API
|
|
url = "https://api.openai.com/v1/responses"
|
|
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
|
|
}
|
|
|
|
async with httpx.AsyncClient(timeout=30) as client:
|
|
r = await client.post(url, headers=headers, json=body)
|
|
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:
|
|
cmd = json.loads(content)
|
|
except Exception as e:
|
|
raise HTTPException(500, f"Failed to parse model JSON: {e}; content={content[:200]}")
|
|
|
|
# Forward to the agent
|
|
async with httpx.AsyncClient(timeout=15) as client:
|
|
r = await client.post(f"{AGENT_URL}/command", json=cmd)
|
|
if r.status_code >= 400:
|
|
raise HTTPException(r.status_code, f"Agent error: {r.text}")
|
|
return r.json()
|