gemini-llm-council

Gemini LLM Council is an extension that lets you query multiple top-tier LLMs in parallel, with automated peer review and synthesis. It aggregates insights from several models, applies personas for targeted reviews (such as security or performance), and provides an audit trail via MCP Resources. Use the council to investigate codebases, verify architectural decisions, and generate high-confidence debugging guidance. You can run one-shot consultations, or use the autonomous investigator to have a subagent read files and deliberate before presenting a consolidated answer.

To use it, first configure the extension with your OpenRouter API key and set up your council members either globally or per-project. Then you can issue commands like /council:setup to choose models and reasoning depth, /council:ask to consult on a query, /council:investigate to have the autonomous subagent explore your project, and /council:persona to apply a specific persona (e.g., security, performance) to the review. The system will automatically ground findings to your project metadata and output a summarized report with a recommended fix or next steps.

Prerequisites:

• Node.js and npm installed on your machine
• Git installed
• Access to the Gemini CLI and an OpenRouter API key

Installation steps:

1. Clone or download the repository for the Gemini LLM Council extension.
2. Install dependencies:

   npm install
   

3. Build the extension:

   npm run build
   

4. Configure your API key for Gemini:

   gemini extensions link .
   gemini extensions config gemini-llm-council "OpenRouter API Key"
   

5. Run or integrate the built extension into your environment as appropriate for your setup (the exact runtime path may vary based on your project structure). If you publish to npm, you can also install and run it via npx as shown in the MCP configuration.

Prerequisites recap:

• Node.js (recommended latest LTS)
• npm
• Git
• Gemini CLI and OpenRouter API key

Tips and notes:

• You can store council configuration globally (in your user home) or project-scoped (in your repository) to control which models and reasoning depth are used.
• Personas are customizable; you can add your own in ~/.gemini/extensions/gemini-llm-council/personas.json.
• For large projects, use the autonomous investigate command to let a subagent read files and gather evidence before delivering a final assessment.
• The results can be accessed via MCP Resources through council:// URIs embedded in the summary report for deep-deliberation trails.
• If you encounter API key or network auth issues, verify environment variables and ensure the Gemini CLI configuration matches your OpenRouter account permissions.