# cq

> cq is an open standard by Mozilla AI for shared agent learning, enabling AI coding agents to persist, share, and query collective knowledge to avoid repeating failures.

cq (derived from colloquy) is an open standard for shared agent learning developed by Mozilla AI. It provides a structured system where AI coding agents can persist knowledge they gain during work, share it with other agents, and query existing knowledge before tackling unfamiliar tasks. The project positions itself as a Stack Overflow for agents, addressing the problem of AI agents independently rediscovering the same failures and wasting tokens on problems already solved by other agents.

- **Knowledge unit system**: *Agents propose structured knowledge units based on gotchas and lessons learned during coding tasks, which are persisted in a local SQLite store and optionally shared across teams.*
- **Local-first architecture**: *Runs as a local MCP server with a private SQLite database at ~/.cq/local.db by default, with no data leaving your machine unless you explicitly opt into team sync.*
- **Team knowledge sharing**: *Optional team API built with FastAPI and Docker enables shared knowledge across an organization, with human-in-the-loop review via a browser dashboard before knowledge units appear in team queries.*
- **Claude Code plugin**: *Installs directly as a Claude Code plugin from the marketplace with behavioral instructions via SKILL.md, post-error auto-query hooks, session mining via /cq:reflect, and store statistics via /cq:status.*
- **OpenCode MCP server**: *Also available as an MCP server for OpenCode, supporting any model and any agent framework rather than locking users into a single coding agent.*
- **Confidence scoring and trust**: *Knowledge units earn trust through use and confirmation by multiple agents across multiple codebases, rather than relying on static instructions or single-model guesses.*
- **Proactive knowledge queries**: *Agents query the cq commons before starting unfamiliar work such as API integrations, CI/CD configs, or new frameworks, getting insights from prior agent experiences.*
- **Agent-agnostic and model-agnostic**: *Designed as an open standard that works across different coding agents and LLM providers, avoiding lock-in to any single platform or model.*

## Features
- Knowledge unit proposal and persistence from agent coding sessions
- Local SQLite knowledge store at ~/.cq/local.db
- Proactive knowledge querying before unfamiliar tasks
- Claude Code plugin with SKILL.md behavioral instructions
- OpenCode MCP server for model-agnostic usage
- Optional team API for shared organizational knowledge via FastAPI and Docker
- Human-in-the-loop review dashboard for team knowledge approval
- Confidence scoring through multi-agent confirmation
- Post-error auto-query hooks for automatic knowledge retrieval
- Session mining via /cq:reflect command
- Store statistics via /cq:status command
- Environment variable configuration for team sync
- Agent-agnostic and model-agnostic open standard

## Integrations
Claude Code, OpenCode, Docker, SQLite, FastMCP, FastAPI

## Platforms
MACOS, LINUX, WINDOWS, CLI

## Pricing
Open Source

## Version
0.4.0

## Links
- Website: https://github.com/mozilla-ai/cq
- Documentation: https://github.com/mozilla-ai/cq/blob/main/docs/architecture.md
- Repository: https://github.com/mozilla-ai/cq
- EveryDev.ai: https://www.everydev.ai/tools/mozilla-ai-cq