The agent never sees your keys
Your agents never hold a real secret.
Comment the release notes on the billing-api pull request.
- write
github · comment_issue
credential attached server-side
The problem with secrets in context
The moment an API key or OAuth token reaches an agent, it lives in the model's context window, in tool transcripts, and often in logs. From there it can be exfiltrated by a prompt injection, leaked in an error message, or simply over-retained. And because those credentials are usually broad and long-lived, a single leak can mean rotating secrets across your whole stack under pressure.
How credential vaulting works
Secrets stay in the vault
Real credentials are stored encrypted in Clawvisor and never leave it. The agent is never issued a credential, or a placeholder standing in for one.
Swapped at call time
When the agent makes a call, the gateway attaches the real secret as the request leaves, so the agent never holds it.
Scoped & short-lived
Every call is bound to the tool and task it was approved under, and the grant expires with the task. There is no credential in the agent's possession to steal or replay.
Rotate without redeploys
Rotate the underlying secret in the vault and every handle keeps working, with no agent code or config to touch.
A leaked agent context exposes no credential at all, let alone your production keys.
It’s one of the power-ups you get from approving the task instead of the tools.
Credential injection
Your API keys and OAuth tokens live in an encrypted vault. Clawvisor injects credentials server-side and returns sanitized results, so a leaked prompt or transcript can't leak a token.
Let your agents act. Safely.
Connect your accounts and put a gatekeeper in front of your agents in minutes. Free to start, no credit card.