agent-vault

Why we included this project

AI agents that call external APIs usually hold the API keys themselves, which is exactly the setup that prompt injection exploits: a cleverly crafted message can trick the agent into handing over a key it was never meant to reveal. Agent Vault removes the key from the agent entirely. Credentials live in a vault, and the agent sends its HTTP requests through a proxy that swaps in the real key at request time, so the agent only ever sees dummy placeholders. It runs as a single binary that serves as both the server and the CLI client, and you can point it at an external secrets store like Infisical instead of the built-in encrypted one. Egress filtering and request logging give you control over which agents can reach which services and let you inspect authenticated traffic, so you can enforce least-privilege access without touching agent code.