opencode-gemini-auth

You can enable "thinking" features per-model using thinkingConfig within provider.google.models. The available fields depend on the model family:

• Gemini 3 models: Use thinkingLevel with values "low" or "high".
• Gemini 2.5 models: Use thinkingBudget (a token count).
• includeThoughts (boolean): Controls whether the model emits internal thoughts.

Example configuration:

{
  "provider": {
    "google": {
      "models": {
        "gemini-3-pro-preview": {
          "options": {
            "thinkingConfig": {
              "thinkingLevel": "high",
              "includeThoughts": true
            }
          }
        },
        "gemini-2.5-flash": {
          "options": {
            "thinkingConfig": {
              "thinkingBudget": 8192,
              "includeThoughts": true
            }
          }
        }
      }
    }
  }
}

To use your Google account's Gemini plan and quotas within Opencode, add the plugin to your Opencode configuration file (typically ~/.config/opencode/opencode.json).

Note: If you are using an organization-backed Gemini Code Assist subscription (Standard or Enterprise), you must explicitly configure a Google Cloud projectId.

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-gemini-auth@latest"]
}

Opencode does not automatically update plugins. To update to the latest version, you must manually clear the cached plugin files and then run Opencode to trigger a fresh installation.

# Clear the specific plugin cache
rm -rf ~/.cache/opencode/node_modules/opencode-gemini-auth

# Run Opencode to trigger a fresh install
opencode

Follow these steps to link your Google account to the Opencode CLI:

1. Run the login command:

   opencode auth login

2. Select Google from the provider list.
3. Select OAuth with Google (Gemini CLI).
4. Approve access in the browser window that opens.

If the local callback server fails (e.g., due to a port conflict or a headless environment), you can manually paste the callback URL or the authorization code when prompted.

You can control which models appear in the Opencode picker using whitelist or blacklist settings under provider.google. Use exact model IDs (verifiable via opencode models google).

• whitelist: Only show the listed model IDs.
• blacklist: Hide specific model IDs from the default list.

{
  "provider": {
    "google": {
      "whitelist": [
        "gemini-2.5-flash",
        "gemini-2.5-pro",
        "gemini-3-flash-preview",
        "gemini-3-pro-preview"
      ]
    }
  }
}

If your network requires an HTTP proxy for Google API calls, set the OPENCODE_GEMINI_AUTH_PROXY environment variable before starting Opencode. This applies to OAuth, token refresh, project/quota lookup, and Gemini request forwarding.

OPENCODE_GEMINI_AUTH_PROXY=http://127.0.0.1:8080 opencode

By default, the plugin attempts to find a suitable project. To force a specific project (required for Standard/Enterprise subscriptions or Workspace accounts), use one of the following methods:

1. Opencode Config File: Set projectId under provider.google.options in ~/.config/opencode/opencode.json.

2. Environment Variables: Set OPENCODE_GEMINI_PROJECT_ID, GOOGLE_CLOUD_PROJECT, or GOOGLE_CLOUD_PROJECT_ID.

{
  "provider": {
    "google": {
      "options": {
        "projectId": "your-specific-project-id"
      }
    }
  }
}

If you encounter 429 RESOURCE_EXHAUSTED or QUOTA_EXHAUSTED errors, consider the following:

• Missing Project ID: Without a configured projectId, the plugin uses a managed free-tier project with lower quotas.
• Model Limits: Quotas are tracked per model (e.g., gemini-3-pro-preview has different limits than gemini-3-flash-preview).
• Large Prompts: OAuth/Code Assist does not support cached content; long system prompts and history consume quota quickly.
• Parallel Sessions: Multiple Opencode windows share the same quota bucket.
• Subscription Type: If using Gemini Code Assist Standard or Enterprise, ensure provider.google.options.projectId is set.

To view detailed logs of Gemini requests and responses, set the OPENCODE_GEMINI_DEBUG environment variable to 1. This will generate gemini-debug-<timestamp>.log files in your working directory containing sanitized request/response details.

OPENCODE_GEMINI_DEBUG=1 opencode

The startOAuthListener function accepts an optional OAuthListenerOptions object:

When interacting with Gemini models that support reasoning, you can provide a ThinkingConfig object. The configuration keys vary depending on the model version:

• Gemini 3 models: Use thinkingLevel with values 'low', 'medium', or 'high'.
• Gemini 2.5 models: Use thinkingBudget (a number).

You can also use includeThoughts (boolean) to control whether the model's reasoning process is returned in the response.