Incomplete Response with useChat in Production (GPT-4o)

bolt.diy [bolt.diy] Issues and Troubleshooting
1

Describe the bug

Issue Summary

I am using useChat from 'ai/react' with OpenAI’s GPT-4o model. In the production version of my app, I am intermittently receiving incomplete responses, but onFinished is still being triggered as if the response were complete.

This issue does not occur in the development version, only in production.

Environment Details

  • Library: ai/react
  • Model: gpt-4o
  • Deployment: Hosted on Cloudflare
  • Issue Occurrence: Only in production, works fine in development

Additional Context

  • This does not happen in development mode.
  • My site is hosted on Cloudflare, which might introduce caching or network-related behaviors.
  • No network errors are reported in the console or logs.

Would appreciate any insights or fixes regarding this issue! Let me know if additional logs or debugging information is required. Thanks! :raised_hands:

Link to the Bolt URL that caused the error

Steps to Reproduce

  1. Use useChat from ai/react to send a request to OpenAI’s GPT-4o.
  2. Deploy the application to production (hosted on Cloudflare).
  3. Observe that in some cases, the response is cut off or incomplete, yet onFinished still fires.

Expected Behavior

  • The full response should be received before onFinished is triggered.
  • If the response is incomplete, onFinished should not fire prematurely.

Screen Recording / Screenshot

No response

Platform

  • OS: Linux
  • Browser: Chrome
  • Version: 132.0.6834.110

Provider Used

OpenAI

Model Used

GPT-4o model