Hacker News digs into Mozilla.ai's cq, a Stack Overflow-style memory layer for coding agents

Hacker News pushed Mozilla.ai’s “Show HN: Cq – Stack Overflow for AI coding agents” to 137 points and 48 comments on March 23, 2026. The idea resonated because it targets a pain point nearly every coding-agent user now recognizes: agents repeatedly hitting the same stale-data mistakes, wasting tokens, and encoding fragile fixes in static repo instructions.

Mozilla.ai presents cq as a local-first memory layer for agents. A plugin for Claude Code and OpenCode talks to a local MCP server backed by SQLite, so knowledge stays on the developer’s machine by default. Teams that want sharing can add an API service and a browser review UI, letting humans approve or reject proposed “knowledge units” before they become reusable guidance for other agents. That keeps the system closer to operational memory than to another giant instruction file.

• Agents query the commons before unfamiliar work and propose new knowledge after discovering something useful.
• Trust is supposed to come from confirmation, reuse, and review rather than from one model’s authority.
• Mozilla.ai is positioning the project as an open, cross-model standard instead of a feature locked to a single vendor workflow.

The HN discussion mattered because the pitch is not “train a better model” but “make everyday agent workflows less forgetful and less wasteful.” Mozilla.ai argues that repo-level Markdown files such as AGENTS.md or CLAUDE.md are too static for narrow, fast-changing gotchas like outdated GitHub Actions versions or API quirks. In that framing, shared memory becomes a living system that can decay, be reviewed, and regain trust over time.

For engineering teams, this is a meaningful design shift. If shared agent memory becomes a standard layer, the competitive advantage may move from prompt tricks to reviewable, portable operational knowledge that works across models. Original sources: Mozilla.ai blog and GitHub repo. Community discussion: Hacker News.