openclaw - 💡(How to fix) Fix [Bug]: Assistant returns audio replies instead of text in dashboard without user request [2 comments, 2 participants]

TL;DR

The issue can likely be mitigated by adjusting the conversation handling after compaction to ensure text responses are prioritized unless audio output is explicitly requested.

Guidance

• Review the conversation compaction logic to identify where the response type is determined and ensure it defaults to text unless the user has requested audio.
• Verify that the gemini flash preview 3 model configuration does not override the response type to audio under certain conditions, such as long conversations.
• Check the OpenClaw version's documentation for any known issues related to response types after compaction and consider updating if a fix is available in a later version.
• Test the scenario with different models or providers to isolate if the issue is specific to the opencode provider or the gemini flash preview 3 model.

Example

No specific code example can be provided without more details on the OpenClaw API or the compaction logic. However, the fix might involve modifying a condition that checks for user input or conversation state to correctly determine the response type.

Notes

The provided information does not include specific logs or error messages, which could help in pinpointing the exact cause. The issue seems to be related to how conversations are handled after compaction, possibly involving the model or provider settings.

Recommendation

Apply a workaround by reviewing and adjusting the conversation compaction logic to prioritize text responses unless audio is explicitly requested, as updating to a potentially fixed version is not explicitly implied in the given information.