Fida is a local-first secret-leak prevention layer for AI coding agents. It stops credentials from ever reaching model context: when an agent reads a .env or runs a command, Fida returns a redacted view, while the real command still executes with the true secret, so the work succeeds. The agent uses a secret without seeing it. Fail-closed by design, nothing leaves your machine. Not a heavy policy firewall — one focused job, done right.