Agent Governance
Agent governance lets external runtimes route consequential steps through Orlo for policy, validation, approval, and audit decisions.
Core concepts
- Agent session — one governed trajectory
- Agent step — one model/tool/retrieval/decision boundary
- Tool policy — allow, warn, reject, or require approval
- Runtime limit — max steps, tool calls, write actions, inactivity timeout
- Approval — human checkpoint for gated steps
Sessions
POST /v1/agent-sessionsGET /v1/agent-sessionsGET /v1/agent-sessions/:session_idPATCH /v1/agent-sessions/:session_id
Create a session before beginning a governed trajectory. Sessions can later be marked completed, failed, or cancelled.
Govern steps
POST /v1/agent-sessions/:session_id/steps/govern
Primary governance endpoint for step evaluation.
Typical step types:
model_callmodel_outputtool_requesttool_resultretrievaldecision
POST /v1/agent-sessions/:session_id/tools/check
Fast path for tool authorization checks.
Trace import and events
POST /v1/agent-sessions/:session_id/traces/importPOST /v1/agent-sessions/:session_id/stepsGET /v1/agent-sessions/:session_id/events
These routes support imported traces, externally recorded steps, and formatted session event views.
Policies and limits
GET /v1/tool-policiesPOST /v1/tool-policiesPUT /v1/tool-policies/:idGET /v1/runtime-limitsPOST /v1/runtime-limitsPUT /v1/runtime-limits/:id
Approvals
GET /v1/agent-approvalsPOST /v1/agent-sessions/:session_id/steps/:step_id/request-approvalPOST /v1/agent-approvals/:id/approvePOST /v1/agent-approvals/:id/reject
Notes
- Orlo is the governance layer, not the orchestration runtime.
- The public API is designed to work well with thin wrappers and callbacks so application developers do not need to wire raw governance calls manually.