Quick Start¶
Get OpenAICE running in under 5 minutes with a replay demo.
1. Run Your First Replay¶
Replay scenarios let you test the full pipeline without any live infrastructure:
python -m openaice.cli.cli replay \
--scenario examples/telemetry-replay/k8s-inference-queue-pressure
This loads synthetic telemetry data simulating a Kubernetes inference service under queue pressure, runs it through the policy engine, and outputs recommendations.
Expected Output
You should see a recommendation to scale_replicas for the inference-api service with confidence 0.91 and risk level medium.
2. Try the Slurm Scenario¶
Expected Output
Recommendations to quarantine_node for the unhealthy GPU node with risk level high.
3. Start the API Server¶
Then visit http://localhost:8000/docs for the interactive OpenAPI documentation.
API Endpoints¶
| Endpoint | Method | Description |
|---|---|---|
/health |
GET | Health check with control/policy mode |
/state |
GET | Current canonical entity state |
/recommendations |
GET | Active recommendations |
/audit |
GET | Audit event history |
/replay |
POST | Run a replay scenario via API |
4. Run a Replay via API¶
curl -X POST http://localhost:8000/replay \
-H "Content-Type: application/json" \
-d '{"scenario_path": "examples/telemetry-replay/k8s-inference-queue-pressure"}'
5. Inspect State¶
python -m openaice.cli.cli state \
--scenario examples/telemetry-replay/k8s-inference-queue-pressure \
--format yaml
This dumps the canonical entity state that the policy engine reasons over.
6. Run Tests¶
All 30 tests should pass, covering schemas, state bus, policy engine, guardrails, and end-to-end replay scenarios.
Next Steps¶
- Configuration Guide — Customize adapters, policy mode, and guardrails
- Architecture Overview — Understand the full pipeline
- Writing Custom Adapters — Integrate your own telemetry sources