Summary

CQ (Colloquy) is an open standard designed to enable shared agent learning by allowing AI agents to persist, share, and query collective knowledge. It addresses the problem of AI agents independently rediscovering the same failures, aiming to improve efficiency and understanding through reciprocal knowledge exchange. The project targets developers building AI systems and agents that benefit from collective experience.

How It Works

CQ operates across three distinct runtime boundaries: the agent process (via Claude Code plugin), a local MCP server (handling knowledge logic and private storage), and a Docker container (providing a team-shared API). Agents interact with the system to find, share, and confirm collective knowledge. The local store uses SQLite, while team synchronization is facilitated by a FastAPI-based API running in a Docker container, enabling shared knowledge across a team.

Quick Start & Requirements

• Claude Code Installation: Use claude plugin install cq or make install-claude from a cloned repository. Requires uv, Claude Code (plugin), and the claude plugin marketplace. Uninstall via claude plugin marketplace remove cq or make uninstall-claude.
• OpenCode Installation: Clone the repository (git clone https://github.com/mozilla-ai/cq.git), navigate into it (cd cq), and run make install-opencode. Requires jq. Project-specific installations are supported. Uninstall via make uninstall-opencode.
• Configuration: Operates in local-only mode by default, storing data in ~/.cq/local.db. Team synchronization requires setting CQ_TEAM_ADDR and CQ_TEAM_API_KEY environment variables, configurable within Claude Code settings or OpenCode configuration files.
• Links: Repository available at https://github.com/mozilla-ai/cq. Architecture details in docs/architecture.md.

Highlighted Details

• Employs a multi-tier architecture involving agent plugins, a local MCP server, and a Dockerized team API.
• Utilizes SQLite for both local (~/.cq/local.db) and team-wide data persistence (/data/team.db).
• Supports agent interaction through commands like /cq:status and /cq:reflect, and post-error auto-query hooks.

Maintenance & Community

No specific details on maintainers, community channels (like Discord/Slack), or roadmap were provided in the README excerpt.

Licensing & Compatibility

Licensed under the Apache 2.0 license. This license is generally permissive for commercial use and integration into closed-source projects.

Limitations & Caveats

The project is explicitly marked as "Exploratory." Further details regarding its proposal and proof-of-concept design are available in the docs/ directory.