#20042·lobehub

[Bug] Conversation content not saved when using Responses API with custom provider (FuguMAX / Sakana AI)

Author: satosi1115jpCreated Sep 27, 2026Updated Sep 27, 2026
Labelsunconfirm

Client Type

Web (Desktop Browser)

Operating System

Windows

Deployment Platform

Official Cloud

Version

LobeHub Cloud

Browser

Firefox

What happened?

Environment

  • Platform: LobeHub Cloud (app.lobehub.com)
  • Model Provider: Custom OpenAI-compatible provider (Sakana AI FuguMAX)
  • Affected Setting: "Use Responses API" enabled

Bug Description

When chatting with FuguMAX on LobeHub Cloud with the Responses API option turned ON, the conversation content (especially assistant messages) is not saved. The chat appears empty after the stream ends.
When Responses API is disabled (Chat Completions mode), conversations are saved correctly.

Steps to Reproduce

  1. Add Sakana AI FuguMAX as a custom OpenAI-compatible provider.
  2. In the model settings, enable "Use Responses API".
  3. Start a new chat and send any message.
  4. Wait for the assistant response to finish streaming.
  5. Reload the page or reopen the topic.
  6. Expected: The conversation history remains.
    Actual: The conversation is empty / not persisted.

Additional Context

I contacted Sakana AI support and received the following technical explanation:

Chat Completions API:
The answer text is returned in choices[].message.content.

Responses API:
The answer text is returned in output[] items with type: "output_text". Specifically, the client should read content[].text from the type: "output_text" item.
It is also important not to treat reasoning items that appear before the answer text as the answer itself.

Sakana AI suggests that the LobeHub client adapter may be referencing a different field instead of the output_text fields above, which matches the reported behavior that it only happens when "Use Responses API" is enabled.

This strongly suggests that the LobeHub adapter for custom providers does not correctly parse output[].type === "output_text" from the Responses API response and therefore fails to persist the assistant's reply.

Possible Cause

LobeHub's custom provider adapter might still expect the Chat Completions response shape (choices[].message.content) even when Responses API is active, causing it to miss the actual text content in output[].


Thank you for looking into this!

How to reproduce it?

No response

What it should be?

No response

Additional Information

No response

️ Willing to Submit a PR?

None

✅ Validations

  • Read the docs.
  • Check that there isn't already an issue that reports the same bug to avoid creating a duplicate.
  • Make sure this is a LobeChat issue and not a third-party library or provider issue.
  • Check that this is a concrete bug. For Q&A, please use GitHub Discussions or join our Discord Server.