Bug: Intermittent "Unexpected response format" error on Gemini models with inconsistent state logging

[Keomars]

Before submitting your bug report

• I've tried using the "Ask AI" feature on the Continue docs site to see if the docs have an answer
• I believe this is a bug. I'll try to join the Continue Discord for questions
• I'm not able to find an open issue that reports the same bug
• I've seen the troubleshooting guide on the Continue Docs

Relevant environment info

OS: macOS Sequoia Apple M4 15.4.1

VS Code Version: 1.103.1 (Universal)

Continue Extension Version: 1.0.23

Model: Gemini-2.5-pro

- config:
  
  provider: gemini
  model: gemini-2.5-pro

Description

Bug Description:
When using a Gemini model, the extension intermittently fails to parse and display a successful API response. This appears to be a client-side parsing issue, as the object logged in the Developer Tools console is a valid, successful response from the Gemini API.

Key Symptoms & Inconsistencies:

Intermittent Nature: The bug does not occur on every request. I have not been able to find a consistent pattern to reproduce it, but it happens with high frequency across different prompts.

UI vs. Payload Discrepancy: Even when selecting Agent or Plan mode in the UI, the "Continue Console" log consistently shows the request Type as Chat. However, inspecting the full Options object in the Continue Console reveals that the model's configuration (e.g., role, tools) does seem to change correctly according to the selected mode.

Conflicting Status Reports: When the error occurs, the main "Developer Tools" console logs the Unexpected response format error. Simultaneously, the "Continue Console" view logs the Result of the same request as Success.

These details strongly suggest potential bugs in the extension's frontend state management, API request construction, and/or response handling logic, specific to the Gemini provider.

Logs:

// From "Developer: Toggle Developer Tools"
[Extension Host] Unexpected response format: {candidates: Array(1), usageMetadata: {…}, modelVersion: 'gemini-2.5-pro', responseId: '...'}

Additional Context:
This issue appears to be separate from the Failed to load LanceDB problem reported in #7192.

To reproduce

1. Configure Environment:

Use VS Code 1.103.1 on macOS with Apple Silicon.

Install the continue extension (v1.0.23).

Configure a Gemini model (e.g., gemini-2.5-pro) as the LLM provider.

2. Perform Action:

Open the Continue side panel.

Select a mode Chat/Plan/Agent.

Send a prompt to the model (e.g., "hello").

3. Observe Inconsistent Behavior:

Symptom 1 (Intermittent Error): The Developer: Toggle Developer Tools console intermittently shows an Unexpected response format error.

Symptom 2 (Incorrect Logged Type): The Continue Console output panel consistently logs the request Type as Chat, even though Agent or Plan was selected.

Symptom 3 (Conflicting Status): When the error occurs, the Developer Tools console shows the error, but the Continue Console logs the Result of the same request as Success with Generated Tokens: 0.

Log output

[Extension Host] Unexpected response format: {candidates: Array(1), usageMetadata: {…}, modelVersion: 'gemini-2.5-pro', responseId: '...'}

[PhungXuanAnh]

I faced the same bug. Does anybody have a workaround for it?

[github-actions]

This issue hasn't been updated in 90 days and will be closed after an additional 10 days without activity. If it's still important, please leave a comment and share any new information that would help us address the issue.

[RomneyDa]

I believe this will be fixed by #8907