[BUG]: Synchronous LLM requests block FastAPI server during form filling

[Mahendrareddy2006]

---

name: 🐛 Bug Report
about: Create a report to help us improve FireForm.
title: "[BUG]: Synchronous LLM requests block FastAPI server during form filling"
labels: bug
assignees: ''

⚡ Describe the Bug

The form filling API endpoint blocks the entire FastAPI server when processing LLM requests.
Since the LLM calls in llm.py use synchronous requests.post() and the route handler in forms.py is synchronous, any delay or failure in Ollama causes the server to hang indefinitely.

👣 Steps to Reproduce

1. Start the FastAPI server
   uvicorn api.main:app --reload

2. Send a POST request to /forms/fill

3. If Ollama is slow or unresponsive, the request hangs.

4. Try making another request to any endpoint.

Result:
The server becomes unresponsive until the first request finishes.

📉 Expected Behavior

The API should process LLM requests asynchronously so that multiple requests can be handled concurrently without blocking the server.

🖥️ Environment

OS: Windows
Python: 3.13.1
FastAPI: 0.104+
Ollama: latest
Model: mistral

🕵️ Possible Fix

The issue appears to be caused by synchronous HTTP requests in llm.py:

response = requests.post(OLLAMA_URL, json=payload)

Potential solution:

Replace requests with httpx.AsyncClient
Convert LLM processing functions to async
Use await for LLM requests
Update FastAPI route handlers accordingly

Example:

import httpx

async with httpx.AsyncClient() as client:
response = await client.post(OLLAMA_URL, json=payload)

This would prevent the FastAPI event loop from being blocked during LLM processing.

[Mahendrareddy2006]

Hi, I noticed this behavior while testing the FireForm API and opened this issue to describe it.

I would be happy to work on a fix for this by converting the LLM request pipeline to use asynchronous HTTP calls if that approach makes sense.

[AIZUFA]

I’ve analyzed the synchronous bottleneck in llm.py and forms.py. The current implementation uses the synchronous requests library, which blocks the FastAPI event loop during long-running LLM calls.

Proposed Fix:
I recommend transitioning to httpx.AsyncClient to handle these calls concurrently. This prevents the server from hanging and allows the API to remain responsive to other requests while waiting for the LLM response.

[AIZUFA]

I’ve included a full refactor of this async architecture in my GSoC proposal to ensure FireForm is production ready for high concurrency environments.

[Mahendrareddy2006]

Thanks for the analysis. I also noticed the blocking behavior while testing locally and it seems to come from the synchronous 'requests.post()' call inside the LLM pipeline.

If the project prefers a larger async orchestration approach, I’d still be happy to contribute by helping transition the LLM request handling to an async implementation as part of that effort.