cq: Share Collective Learning Between AI Agents

Overview

cq gives AI coding agents a shared store of structured lessons, so an agent can check for relevant guidance before tackling a problem and contribute useful discoveries afterward. It is designed to reduce repeated debugging and make knowledge from one agent session useful in others.

Knowledge can remain in a local SQLite store or be shared through Mozilla.ai's hosted service or a self-hosted server. The project is in 0.x, so breaking changes are expected.

Key Features

• Provides MCP tools to query, propose, confirm, flag, and check the status of knowledge units.
• Guides agents to search existing knowledge before acting or retrying after an error.
• Supports optional session reflection to identify lessons for review and submission.
• Offers local storage, a hosted remote service, and self-hosted server deployment.
• Includes installation targets for Claude, Codex, Copilot, Cursor, Devin Desktop, OpenCode, and Pi.
• Publishes a CLI, Go and Python SDKs, a schema, and a server image.

Use Cases

• Individual developers can reuse solutions to recurring errors across coding-agent sessions.
• Teams can run a shared server so agents and developers can access a common pool of reviewed lessons.
• Developers using multiple coding-agent hosts can make a knowledge store available through supported integrations.
• Agent maintainers can capture undocumented API behavior and non-obvious workarounds for future tasks.

Getting Started

Install the CLI, then install cq into a supported coding agent:

cq install --target <host>

See the repository README and documentation for installation details, quickstart instructions, and remote server setup.

Alternatives

• byterover-cli: ByteRover CLI stores and queries curated project context locally with optional cloud sync, while cq focuses on reusable coding lessons and validating contributions.
• agentmemory: agentmemory provides shared persistent memory across sessions via MCP and REST, while cq centers on querying and validating reusable lessons.
• ReMe: ReMe turns conversations and resources into evolving Markdown memory, while cq organizes reusable coding lessons with a contribution and validation workflow.

Considerations

• The project is in 0.x status and warns that breaking changes should be expected.
• Remote sharing requires configuring a service URL; write operations require an API key. Local use does not require a remote service.
• Self-hosting means deploying and operating the separate server component.
• Knowledge proposed locally is sent to the remote store when the plugin starts and becomes available to agents after human review.