Skip to main content
This page is for enterprise and startup teams running agent workloads in production. At request time, answer two questions:
  1. Is this request really from the claimed agent?
  2. Is this agent allowed to perform this action right now?

Auth quick reference

Request-time flow

Step 1: verify identity

  • Option A (recommended): local Ed25519 verification
  • Option B: online verification with POST /v1/verify
Example response:

Step 2: enforce policy

Call POST /v1/check with agent_id, policy_id, action, and optional context.
Example response:
Possible decisions: allow, challenge, deny.

Control-plane flow

Agent inventory mapping

  • POST /v1/workspace/agents/upsert
  • GET /v1/workspace/agents

Audit query and export

  • GET /v1/workspace/audit/events
  • POST /v1/workspace/audit/export

Short-lived credentials

  • POST /v1/identity/credentials/mint
  • POST /v1/identity/credentials/revoke
  • POST /v1/identity/credentials/revoke-by-agent

Incident controls

  • POST /v1/identity/kill-switch/minting
  • POST /v1/identity/kill-switch/revoke-all
  • GET /v1/identity/kill-switch/status
  1. Edge layer verifies inbound agent signatures.
  2. Decision layer calls /v1/check right before execution.
  3. Control-plane jobs sync inventory and export audits.
  4. Sensitive follow-on actions use short-lived credentials or Vault proxy.
  5. Security operations can pause minting/revoke quickly during incidents.

Next steps