gemini-cli - 💡(How to fix) Fix Voice mode cloud backend requires GEMINI_API_KEY and does not support existing CLI auth [2 comments, 3 participants]

Other CLI features work transparently for OAuth users via the existing content generator/auth path; voice mode being the exception is surprising.

TL;DR

The issue can be fixed by adding a fallback for OAuth users to use their existing authentication for cloud voice mode without requiring a separate GEMINI_API_KEY.

Guidance

• Investigate modifying useVoiceMode.ts to use the existing content generator/auth path for OAuth users, similar to other CLI features.
• Consider adding a conditional check to use the OAuth authentication token when GEMINI_API_KEY is not set.
• Review the transcriptionFactory.ts to determine the best way to integrate the new authentication path without disrupting the existing Live API path.
• Evaluate the feasibility of adding an optional authentication parameter to the Live API WebSocket URL to support both API key and OAuth authentication.

Example

// Example of conditional check for OAuth authentication token
const apiKey = config.getContentGeneratorConfig()?.apiKey || process.env['GEMINI_API_KEY'];
if (!apiKey && oauthToken) {
  // Use OAuth token for authentication
  const wsUrl = `wss://...?token=${oauthToken}`;
} else {
  // Use API key for authentication
  const wsUrl = `wss://...?key=${apiKey}`;
}

Notes

The proposed solution assumes that the OAuth authentication token can be used as a replacement for the GEMINI_API_KEY in the Live API WebSocket URL. Further investigation is needed to confirm this and determine the exact implementation details.

Recommendation

Apply workaround by adding a fallback for OAuth users to use their existing authentication for cloud voice mode, as this is a more immediate solution that can be implemented without disrupting the existing Live API path.