Onbe documentation

Onbe is an agent identity and delegation control plane. It issues short-lived, scoped tokens that let an AI agent act on behalf of a specific human, and keeps a hash-chained audit trail that proves which human.

Everything here is standards-based: RFC 8693 for token exchange, RFC 7662 for introspection, RFC 7009 for revocation and RFC 8707 for resource indicators. Agents authenticate with private_key_jwt, signing a short-lived assertion with their own key. There is no client_secret_post and there will not be one: static shared secrets are the thing the project exists to remove.

Start here

What Onbe is, and what it is notThe problem it solves, the problems it does not, and what it expects you already run.

QuickstartA control plane running and a token exchanged, from nothing, on a clean machine.

Your first exchangeThe request, the response, and every field in the token it hands back.

The five invariantsThe rules the server keeps, each one guarded by a conformance test.

Understand it

Delegation, not impersonationWhy the subject stays the human and the agent lives in act.

Tasks and task grantsWhy a six-hour job never holds a six-hour credential.

ScopeAn intersection of three sets, and why it only ever gets smaller.

Delegation depthWhat a chain of agents means to the tool server at the end of it.

RevocationEventual and cheap, or immediate and a round trip. The tradeoff, stated plainly.

The audit ledgerWhat the hash chain actually proves, and what it does not.

Build with it

Register an agent

Build an agent

Protect a tool server

Protect an MCP server

Operate it

SDK reference

Trust it

Threat modelWhat Onbe defends against, and what is out of scope for v0.1.

The conformance suiteTwelve adversarial tests run against a live instance through its public surface.

Reporting a vulnerabilityWhere to send it and what to expect back.