POST /api/graph/ontology/generate HTTP/1.1" 500

[girishhc-cs]

I am trying to run it locally on my laptop but i am getting 500 error

[dosubot]

Hi @girishhc-cs! I'm Dosu and I’m helping the MiroFish team.

This is a frequently reported issue. The most common cause of the 500 error on POST /api/graph/ontology/generate is that the chat_json() method unconditionally sends response_format={"type": "json_object"} to your LLM provider, which many providers (OpenAI, Groq, SiliconFlow, etc.) either reject or return empty content for — causing a JSON parse failure and 500 error. MiroFish is primarily tested with Alibaba Bailian qwen-plus [1].

To fix this, apply the fallback retry patch to backend/app/utils/llm_client.py (from PR #186):

def chat_json(self, messages, temperature=0.3, max_tokens=4096):
    try:
        response = self.chat(messages=messages, temperature=temperature,
                             max_tokens=max_tokens,
                             response_format={"type": "json_object"})
    except Exception:
        # Retry without response_format for unsupported providers
        response = self.chat(messages=messages, temperature=temperature,
                             max_tokens=max_tokens)
    # ... rest of existing JSON cleaning logic

After patching, clear cached bytecode and restart:

find backend -type d -name __pycache__ -exec rm -rf {} +
PYTHONDONTWRITEBYTECODE=1 npm run dev

Other things to check:

• Ensure LLM_BASE_URL ends exactly at /v1 (e.g. https://api.openai.com/v1) [2]
• Verify your LLM_API_KEY is valid — test it with a direct API call first
• The 500 response body includes a traceback field [3] — check it in browser DevTools (Network tab → failed request → Response) for the exact error

Related issues: #271, #298, #299 — all resolved with the same fix.

Could you share which LLM provider/model you're using and the traceback from the 500 response? That would help pinpoint the exact cause.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

  

[LBIAORNCELL1007]

Just use grok it works just as fine. Tried and tested it