What is AgentValet?
AgentValet is a governed credential broker for AI agents. When an agent — Claude, a CLI script, an internal automation — needs to send an email, post to Slack, charge a card, or read a row from your CRM, it doesn’t get your API keys. It calls AgentValet, and AgentValet decides whether to let it through.
The problem
Agents are getting more capable. They can install software, talk to APIs, move money. The way most teams give them access today is to paste a long-lived API key into a config file and hope nothing goes wrong.
That’s a bad trade. One leaked key is a full-organisation compromise. One mistaken prompt can drain a Stripe balance before anyone notices.
What AgentValet does
- Issues each agent its own identity. A cryptographic keypair, not a shared secret. You can see exactly which agent took which action.
- Holds your platform credentials in a vault. Agents never see them. Credentials are decrypted in-memory only at call time, never logged.
- Enforces a deny-by-default permission matrix. An agent gets exactly the scopes you grant — nothing more.
- Queues actions for human approval when you ask it to. Every platform grant carries a
require_approvaltoggle — flip it on and every call under that grant, reads included, queues for you to decide. Policies can also mark individual actions as ask first. - Records every action in an audit log. Including the requests that were denied.
- Trips a circuit breaker on repeated failures. Three failed credential refreshes or five upstream failures within ten minutes block the agent’s calls; after a five-minute cooldown, one successful call closes the breaker again.
Who it’s for
- Solo builders running personal agents who don’t want their email/CRM/payment keys living in plaintext on disk.
- Teams that want one place to revoke an agent’s access — across every platform it touches — when an employee leaves or a script goes rogue.
- Compliance-sensitive orgs that need a record of what an autonomous system did and why.
How it fits in
You connect your platforms once. You register agents and grant them scopes. From then on, everything an agent does goes through AgentValet — observable, governed, revocable.