[BUG] TypeError: fetch failed sending request for any task with Gemini

[evil-shrike]

Problem (one or two sentences)

Sending any request (with ANY prompt) to ANY Gemini model fails with this generic error:
API Request Failed
Gemini generate context stream error: exception TypeError: fetch failed sending request

Initially reported here:
#9465

first version where it appears: 3.33.2
last checked version where it's actual: 3.38.3

Context (who is affected and when)

The issue doesn't seem to be related to user's actions, any task via Gemini fails

Reproduction steps

select Gemini with any model
provide any text for a task

Expected result

response from Gemini

Actual result

A generic error like the API endpoint isn't reachable

Variations tried (optional)

No response

App Version

3.38.3

API Provider (optional)

Google Gemini

Model Used (optional)

No response

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote]

I have analyzed this issue and the related issue #9465. Here is my proposed implementation plan:

Analysis

The error "TypeError: fetch failed sending request" is a generic wrapper from the @google/genai SDK that obscures the underlying cause. This could be due to:

• Proxy/firewall configurations not being respected by native Node.js fetch
• Request timeouts (Node.js fetch has a default 300s timeout)
• Network connectivity issues specific to the extension context

The issue started appearing in v3.33.2, but the SDK version (@google/genai ^1.29.1) has not changed. The underlying @google/genai SDK has known issues with error handling (see googleapis/js-genai#1001).

Proposed Plan

1. Improve error capture and logging: Wrap the Gemini API calls to capture the original error cause before the SDK wraps it in a generic message. This will help users diagnose the actual issue (timeout, DNS, SSL, proxy, etc.)

2. Add timeout configuration: Pass a configurable timeout value via httpOptions.timeout to prevent silent failures on long-running requests

3. Add retry logic with exponential backoff: Implement automatic retries for transient fetch failures (network blips, temporary DNS issues)

4. Add diagnostic logging: Log the actual fetch error details to help users troubleshoot environment-specific issues

This approach will both improve the user experience when errors occur (better error messages) and handle transient failures automatically (retry logic).

Please mention @roomote in a comment saying that you approve this plan, and I will proceed with the implementation.

[hannesrudolph]

@evil-shrike TY for this. I will look into it more in depth tomorrow.