Cq: Building a Knowledge Commons for AI Agents

Mozilla.ai is exploring how to create shared learning infrastructure for AI coding agents with Cq, a new open-source project. Inspired by the success of platforms like Stack Overflow, Cq aims to be a central repository where agents can propose and verify ‘knowledge units’ (KUs) — specific insights or gotchas encountered while using various models.

The core idea is that as agents interact with codebases and tools, they often run into predictable issues. Instead of each agent needing to rediscover these solutions independently, Cq provides a mechanism for them to share what they’ve learned. When an agent encounters a problem, it can propose a KU detailing the issue and solution, which then undergoes peer review (initially by humans) before being added to the knowledge base.

Key Features:

• Standardized Knowledge Units: KUs follow a structured schema, ensuring consistency across different models and applications
• Local-First Design: Data is stored locally by default, with optional team sync capabilities for collaboration
• Human-in-the-Loop Review: A web dashboard allows users to review and approve proposed knowledge units
• Integration Options: Available as a Claude Code plugin or OpenCode MCP server
• Open Source: Licensed under Apache 2.0, encouraging community contributions

Early results are promising, with Cq helping agents avoid common pitfalls like using outdated dependencies or misinterpreting API documentation. One example involved an agent consistently recommending older versions of GitHub Actions until it was taught to check for the latest releases — this insight is now shared via Cq, preventing similar errors in other projects.

While challenges remain around data governance and privacy, the Mozilla.ai team is focused on building a practical tool that developers can use immediately. Platforms like Cq represent an important step toward creating more reliable and efficient AI coding assistants.