Show GN: HTTP credential proxy and secret vault for AI agents
Key point
Infisical unveils an open source HTTP credential proxy and vault for AI agents
Details
Agent Vault is an open source HTTP credential proxy built so that AI agents never see secrets directly.
It intercepts the agent's outbound HTTP requests through a local forward proxy, injects credentials stored in the vault, and forwards them to the target service. It's provided as a portable binary that doubles as a server/CLI client, and can also be deployed via Docker.
The core components are as follows.
- vault: a logical container that holds credentials
- credential: sensitive values such as API keys, passwords, DB credentials
- service: allowed target hosts like
api.stripe.comand their authentication method - agent: agents such as Claude Code, OpenClaw, Hermes
For setup, you point the proxy using HTTPS_PROXY and a trusted CA certificate, and block direct outbound traffic at the network level so that all outbound traffic goes through Agent Vault. This makes it possible to consistently enforce allowlist, rate limiting, audit logging, and per-agent scoping regardless of the interface.
It is currently in the research preview stage, and self-hosting, integration experience, security, and scalability are being continuously refined.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.