Chatbot frontend became unresponsive after receiving a complete LLM response

[siddhantgithub]

PROBLEM (Production Only)

The chatbot frontend became unresponsive after receiving a complete LLM response in production only. Local development worked perfectly.

Symptoms:

• After streaming completed, the app remained stuck in "streaming" state
• The useChat hook's onFinish callback was never triggered
• UI remained in loading state (isLoading: true) indefinitely
• Users could not send subsequent messages
• Input field remained disabled

ROOT CAUSE

The Connection: keep-alive HTTP header in the SSE streaming response prevented proper stream completion in production.

Why it failed in production:

• Vercel's Edge Network (HTTP/2) strictly honors Connection: keep-alive
• The connection remained open indefinitely even after sending [DONE]
• The useChat hook waits for connection close to trigger onFinish
• Without connection close, the stream never completed

Why it worked locally:

• Local development server (uvicorn) closes connections immediately after the generator completes
• Ignores or overrides the Connection: keep-alive directive
• No edge network buffering or HTTP/2 connection pooling

PROPOSED SOLUTION

Remove the Connection: keep-alive header to allow natural connection closure.

Code Change:

File: api/utils/stream.py (Line 252)

Before:
response.headers["Connection"] = "keep-alive"

After:
(Remove this line entirely)