Common Ground is an open-source application designed for building, observing, and collaborating with teams of AI agents. It targets users who need to tackle complex research and analysis tasks through multi-agent systems, offering a transparent and manageable workflow with a focus on human-AI collaboration.

How It Works

The system employs a Partner-Principal-Associate agent architecture, simulating a consulting team. A Partner agent handles user interaction, a Principal agent manages planning and task decomposition, and Associate agents execute specific tasks in parallel. This approach, combined with declarative agent design via YAML files and a unified Model Context Protocol (MCP) for tool integration, allows for flexible customization and deep observability into agent decision-making.

Quick Start & Requirements

• Recommended Deployment: Docker.
• Prerequisites: Docker, Docker Compose, Git. Google Cloud Project ID is required for Gemini OAuth authentication.
• Setup: Clone repo, initialize submodules, navigate to deployment, run docker compose up. Authentication via docker compose run --rm --service-ports login or .env file for API key.
• Access: http://localhost:8000
• Developer Setup: Python 3.12+, Node.js. Backend: cd core; uv venv; uv sync; uv run run_server.py. Frontend: cd frontend; npm install; npm run dev.
• Links: Architecture Documentation

Highlighted Details

• Model-agnostic via LiteLLM, with Gemini pre-configured.
• Full observability with Flow, Kanban, and Timeline views.
• Declarative agent design using YAML for no-code/low-code customization.
• Built-in project and knowledge management with RAG.

Maintenance & Community

• Recent updates (July 2025) focus on core logic refinements, context leak fixes, and UI enhancements.
• Community: Discord server available for support and discussion.
• Contribution: Welcome via Contribution Guide and GitHub Issues.

Licensing & Compatibility

• Licensed under Apache-2.0.
• Permits commercial use and linking with closed-source projects.

Limitations & Caveats

The project is at version 0.1.2, indicating early-stage development. While recent updates address critical bugs like context leaking, users should anticipate potential for further changes and ongoing refinement.