Mozilla AI has released cq, an open-source project designed as a shared knowledge commons for AI coding agents, allowing agents to query prior learnings from other agents and contribute discoveries back to a collective store.
The announcement, published on the Mozilla AI blog, frames the project against the decline of Stack Overflow. According to the post, Stack Overflow peaked at over 200,000 questions per month in 2014 and had fallen to 3,862 questions in December 2025 — back to its launch-month numbers — with the drop tracking closely to the release of large language models. The Mozilla AI post attributes the decline partly to agents and LLMs consuming the platform’s corpus while simultaneously hollowing out the communities that produced it, describing the dynamic as “matriphagy” — offspring consuming the parent.
The practical problem cq aims to address is agents repeatedly solving the same problems independently. According to the post, before an agent tackles unfamiliar work — an API integration, a CI/CD configuration, a framework it has not encountered — it queries the cq commons. “If another agent has already learned that, say, Stripe returns 200 with an error body for rate-limited requests, your agent knows that before writing a single line of code,” the post states. When an agent discovers something novel, it proposes the knowledge back to the commons, where other agents can confirm or flag it as stale.
Mozilla AI describes the cost of the status quo as agents “reading files, writing code that doesn’t work, triggering CI builds that fail, diagnosing the issue, then starting over,” each one burning tokens and compute on the same wall independently.
The system is already past the proof-of-concept stage. Mozilla AI says cq includes a plugin for Claude Code and OpenCode, an MCP server managing a local knowledge store, a team API for sharing knowledge across an organisation, a human-in-the-loop review UI, and containers for spinning the whole stack up. The project is open source and the code is publicly available.
The post cites a data point from developer surveys: 84 percent of developers now use or plan to use AI tools, while 46 percent say they do not trust the accuracy of the output, up from 31 percent the year before. Mozilla AI argues that knowledge confirmed by multiple agents across multiple codebases carries more weight than a single model’s best guess, and that cq’s confirmation layer can address some of that trust gap.
Mozilla AI began building cq at the start of March 2026. Shortly after, the post notes, AI researcher Andrew Ng published a post asking whether there should be a Stack Overflow for AI coding agents, which Mozilla AI describes as confirmation of the direction.
The project explicitly positions itself against lock-in. The post states that not all engineers use the same coding agent, and that mandating a single workflow would repeat what it sees as a mistake in how organisations currently handle AI-assisted development. The current widespread approach — updating Markdown files in repositories and hoping agents adhere — “only gets you so far,” according to the post.
Mozilla AI says it is dogfooding cq internally across its own projects, using it day-to-day to build up knowledge units, identify friction, and determine what matters when agents share knowledge in practice. The team describes the feedback loops cq creates as capable of surfacing patterns across teams and gaps in tooling that only become visible at scale.
The post closes with an invitation for community input on the repo and a published proposal for the knowledge-sharing standard cq is designed to establish.