Capybarrier

An adaptive immune system for AI agents. It learns from every attack, rewrites its own defenses, and ships a fix only with receipts. Capybarrier is a self improving security harness for AI agents. A runbook cache in MongoDB Atlas recognizes known attacks with Atlas Vector Search before any model call, and an evolver agent rewrites the agent's own rules, guardrails, tool access and context policy after every round of attacks. A change ships only with receipts: the exact attacks that got through are replayed against it together with every normal request, and it must fix breaches without breaking anything. Our demo patient is Kapi, the concierge agent of CapySpa (a capybara hot spring resort), running live on Atlas Agent Engine.