Runtime Runs API
REST and stream interfaces for orchestrated runs, approvals, artifacts, and Omar loop execution.
- runtime-api
- sse
- websocket
- omar-loop
- approvals
Runtime APIs expose deterministic run lifecycle and replayable evidence.
Core endpoints
- `POST /api/v1/runs` create run
- `GET /api/v1/runs/{run_id}/status` fetch status and summaries
- `POST /api/v1/runs/{run_id}/cancel` cancel run
- `GET /api/v1/runs/{run_id}/events` SSE event stream
- `GET /api/v1/runs/{run_id}/events/list` paged poll endpoint for deterministic replay
- `WS /api/v1/runs/{run_id}/terminal` terminal stream with resume cursor
- `POST /api/v1/runs/{run_id}/approvals` approve or deny gated actions
- `GET /api/v1/runs/{run_id}/evidence` export run evidence bundle
- `GET /api/v1/runtime/kpi?compare_window_days=&baseline_window_days=` compare and baseline KPI windows
- `POST /api/v1/runs/git/pr/checkpoints` execute git/PR checkpoints across multiple run IDs
- `GET /api/v1/models/catalog` account-aware model catalog and ranking
- `POST /api/v1/loops/omar` start remediation loop
- `GET /api/v1/loops/omar/{loop_id}` inspect loop state and exit reason
Event contract
All runtime events include:
- stable identifiers and timestamps
- event kind, actor, and normalized payload
- usage/accounting metadata for audit reporting
Typical event families
- lifecycle transitions (start, progress, complete, failure)
- tool execution summaries (call + result)
- output streaming updates (terminal and narrative)
- governance checkpoints (approval requested, approved, denied)
Omar loop stop semantics
Loop terminates only when one of the following is true:
- clean `P0-P2` and green quality gates
- explicit budget or iteration ceiling reached
- policy or approval checkpoint denies continuation
Structured Answers
How do I stream run output in real time?
Subscribe to SSE events at /api/v1/runs/{run_id}/events and open terminal WebSocket at /api/v1/runs/{run_id}/terminal for stdout/stderr chunks.
What should I use for audit exports?
Use /api/v1/runs/{run_id}/evidence for downloadable run evidence with timeline, integrity chain metadata, redaction status, findings summary, and artifacts.