Built · control planeSchema 1.0
Integrations and keys
Every external agent is a registered agent bound to one protected account and policy. Its API keys are its credentials: scoped, rotatable and revocable.
Register
In Console → Agents, register the agent with its smart account, chain and policy. Optionally record its framework. The name is a label, never an identity: the key authenticates.
Scoped keys
HTTP
POST /api/v1/agents/{agent_id}/credentials (console session, admin)
{ "label": "production", "scopes": ["operations:write", "operations:read", "policy:read"] }
→ 201 { "id": "…", "prefix": "ps_test_1a2b3c4d", "scopes": [...], "secret": "ps_test_…" } # shown onceoperations:write: authorize, dry-run, withdraw its own pending approvals.operations:read: read its own operations.policy:read: read its bound policy version.audit:read: read its own audit events.mcp: connect through the MCP server.
Keys created before scopes existed hold all of them, so existing integrations keep working. A key can never change its own scopes.
Rotate and revoke
POST /api/v1/credentials/{id}/rotate issues a new secret with the same agent, label and scopes and revokes the old key in the same database statement. Revocation takes effect on the next request. Both are audited.
What a key can read about itself
HTTP
GET /api/v1/integration any valid key: agent, account, chain, scopes, bound policy
GET /api/v1/integration/policy policy:read the exact active policy version and document
GET /api/v1/integration/operations operations:read its own operations (no calldata)
GET /api/v1/integration/audit-events audit:read audit events it caused