Three kinds of principal
Tokens carry a human, delegated-agent, or workload-agent profile. A change made by an agent is attributed to the agent and to the person it acted for.
PlatformAgentLabs Auth
Identity for the agentic era.
AgentLabs Auth is the identity and access control plane for every AgentLabs system. It is built on the premise that the principals in an enterprise are now people, services, and agents, and that an agent's authority should be delegated, scoped, and auditable rather than inherited from a login.
AgentLabs Auth
- Tenant identity
- People, services, and agents in one directory.
- Roles and policies
- Defined once, enforced deny-by-default.
- Access audit
- Every access decision on the record.
Where it stands
AgentLabs Auth is in development. It is the identity provider for every AgentLabs system today, and its releases are under acceptance. It is not offered for production use outside AgentLabs.
Where it is heading
Direction, not shipped capability
Authority borrowed from a named person, scoped to a purpose, narrowing as it passes between agents, and expiring with the task. The delegated-agent profile and policy bundles are the foundation for it.
What it does
One identity authority for people, services, and agents.
- Multitenant identity with memberships, invitations, verified domains, and per-tenant settings.
- An OpenID Connect provider with PKCE, consent, refresh rotation, device flow for CLIs and agents, and enterprise single sign-on through upstream connections.
- People sign in with passkeys, step up for sensitive actions, and administrators can see and revoke every active session.
- Authorisation is defined once and enforced everywhere: each application registers its permissions with Auth, then enforces them locally, deny by default.
- A per-tenant security audit trail and an administration console that keeps tokens out of the browser.
What makes it agent-era
Authority is delegated, never inherited.
Workload authority
Services and agents obtain audience-bound credentials of their own, so agent-to-service calls never borrow a person's session.
Applications bring their own permissions
Each application registers its permission vocabulary and policy bundles with Auth. Every application then enforces them locally through one shared, default-deny evaluator that fails closed.
Built for agents to read
Device flow for agents that cannot hold a browser session, and documentation published in a form agents can consume directly.
The next step
Talk with AgentLabs about AgentLabs Auth.
Bring your identity, policy, and evidence requirements. We will say plainly what is verified and what is not.