Agent systems / Coming soon

GitHub Custodian

GitHub permissions for teams working with AI agents. Scope each action to an explicit policy and keep a signed receipt, without handing agents your personal GitHub credentials.

Agent systems

Let agents work. Keep authority yours.

Giving an agent useful GitHub access should not mean giving it your personal credentials and hoping for the best. GitHub Custodian separates the work an agent wants to do from the authority it needs, with explicit policy and signed receipts for the actions it handles.

Made for

For teams introducing agents into repository workflows who want a clearer answer to three questions: what was allowed, who had authority, and what record remains afterward.

Try it / follow along

Coming soon

We’re still building. Get in touch if you’d like to collaborate or hear more.

Talk about GitHub Custodian

A closer look / the workflow

From the first step
to something useful.

Define the allowed work
Start with a policy describing the GitHub action and its scope. Make the boundary specific enough to assess before the operation happens.
Keep authority separate
Let the agent request work through the custodian's permission model, rather than sharing a person's general-purpose GitHub credentials with the agent.
Keep the receipt
Retain a signed record of the handled action so a later review has something concrete to inspect beyond the agent's own account.

Inside GitHub Custodian

The details
make the difference.

Scoped GitHub permissions

Authority is tied to an explicit policy, giving repository operations a more deliberate boundary than an unrestricted credential.

Signed action receipts

Receipts make the permission story inspectable after the fact. They give a later review a concrete record of the permissioned action to examine.

Agent and person separation

The product is built around preserving the distinction between agent authority and personal access, a useful foundation for teams adopting automated workflows.

Where it stands