I built a trust firewall for my AI agent's memory — on Cognee's four verbs

Built for the WeMakeDevs × Cognee hackathon — "The Hangover Part AI: Where's My Context?"

AI coding agents are finally getting long-term memory. That's the good news. The bad news is the part nobody likes to say out loud:

A memory layer is only as trustworthy as the worst fact in it.

The moment an agent can remember, it can also remember wrong — and confidently hand that wrong thing to the next agent in line. A stale deploy command. A contradicted API contract. An AWS key someone pasted into a note six months ago. Once it's "memory," every future agent treats it as truth.

ContextFirewall is one small idea taken seriously: audit every remembered fact before it reaches the next agent. And because the agents people actually use speak the Model Context Protocol (MCP), I shipped it as an MCP server. Point Claude Code, Cursor, or Windsurf at one endpoint, and from then on every memory the agent recalls, stores, distils, or forgets flows through Cognee and four firewall checks first.