Integrations
Michelangelo runs alongside the ML infrastructure your organization already has. This page is the operator reference for connecting external systems and extending Michelangelo's built-in components. It targets platform engineers responsible for running Michelangelo in production.
Built-in Components
These guides cover components that ship with Michelangelo. Operators configure them; they are not external systems.
| Guide | Description |
|---|---|
| Model Registry | Verify the registry is healthy, configure object store and RBAC, and integrate registered models with serving and CI/CD |
External System Integrations
These guides cover connecting Michelangelo to systems your organization already runs.
| Guide | Description |
|---|---|
| Experiment Tracking | Expose an external experiment tracking server to task pods — network setup, URI injection, and operator/user boundary |
Extending Built-in Components
Michelangelo exposes extension points for replacing or augmenting its core subsystems. Use these when the defaults don't fit your infrastructure.
| Guide | Description |
|---|---|
| Custom Serving Backend | Add support for any inference framework — Triton, vLLM, TensorRT-LLM, or your own |
| Extend the Job Scheduler | Replace or extend the scheduler — integrate Kueue, Volcano, or implement a custom JobQueue and AssignmentStrategy |
| Register a Compute Cluster | Connect an existing Kubernetes cluster so Michelangelo can dispatch Ray jobs to it |
Next Steps
- Platform Setup — ConfigMap reference for all components
- Authentication — OIDC, RBAC, and service-to-service auth
- Network & Ingress — Ingress setup, Envoy proxy config, TLS, multi-cluster networking
- Monitoring — Prometheus metrics, alerting, Grafana dashboards
- Troubleshooting — Common failure modes and
kubectldiagnostic commands