IdP-native
Integrates with Okta and Entra ID. Agent permissions derive from the humans and teams they act for.
Recorded with the runHuman, team, and agent IDs
Infrastructure
Agent Identity
Agents as first-class principals.
Issue every agent its own identity: short-lived credentials, permissions inherited from your IdP, and policy that answers what an agent may do before it acts. No shared service accounts, no standing secrets.
Agent Identity
Example workflow
Requester
Maya Chen
Finance operations
Policy
Refund review
≤ $25k · approval required
Credential
Scoped token
2 resources · 14m remaining
ActionDecisionRule
Read case filesAllowedfinance.read
Create draft refundAllowedrefund.draft
Issue paymentNeeds reviewhuman.approval
How it works
From request to a reviewable result.
Each run keeps the request, actions, evidence, and final decision together.
01 · Resolve
Identify who the agent represents.
Context binds the run to an agent, the requesting person, and the team policy inherited from your IdP.
Output · Principal resolved
02 · Authorize
Evaluate the proposed action.
Policy checks the resource, action, purpose, and approval state before a credential is issued.
Output · Allow, deny, or review
03 · Expire
End access with the task.
The scoped credential expires at completion and its grants and denials remain attached to the run.
Output · Authorization record
What you can inspect
Capability is only useful when it leaves evidence.
Short-lived, scoped credentials
Per-task tokens are minted at run time and expire at completion, reducing reliance on standing credentials.
Recorded with the runScope and expiry
Policy before action
Declarative rules evaluated on every tool call, not after the fact. Deny by default, approve by exception.
Recorded with the runMatched policy rule
Per-action authorization trail
Every grant and denial is logged with the identity, task, and rule that decided it.
Recorded with the runDecision and rationale
Deployment boundaries
Control is part of the product.
Agent Identity can run as a standalone surface, while its traces, evals, and policy decisions remain part of the same Context system.
Connected systems still decide
Context narrows and brokers access; the destination system remains the final enforcement point.
Policy needs ownership
Security teams define allowed resources, approval paths, token lifetime, and emergency revocation.