How we stopped giving our AI agents raw API keys

Autonomous agents need API access to do useful work. Our creature Secure files security issues on GitHub. The voyager genome commits code. Future creatures will need Stripe, analytics, whatever. The naive solution is to inject API keys as environment variables. Every container runtime supports it, every SDK can read from process.env , and it works on day one. It also means every creature has every key, there's no audit trail, and a prompt injection can exfiltrate credentials in a single tool call. We needed something better. Janee: a credential proxy for agents Janee is an MCP server that sits between agents and APIs. You store your credentials in Janee (encrypted at rest with AES-256-GCM), define capabilities with access policies, and agents call APIs by capability name. They never see raw keys. ┌──────────┐ MCP/HTTP ┌────────┐ real creds ┌──────────┐ │ Creature │ ──────────────> │ Janee │ ──────────────> │ External │ │ │ │ │ proxied req │ API │ └──────────┘ └────────┘ └──────────┘ no