Deploy and operate¶
Use this section when you are ready to run DeltaLLM outside a developer's computer or need to keep an existing deployment healthy.
Choose where to run it¶
| Option | Best for |
|---|---|
| Railway | A quick hosted evaluation |
| Docker | Local or single-host evaluation |
| Kubernetes | Evaluation or production on a Kubernetes cluster |
| Gateway ingress admission | Set request and upload limits before authentication |
| Accounting protocol v2 | Set durable budget reservations and batched accounting |
The Docker Compose ha profile is useful for testing more than one DeltaLLM process, but it is not
a production high-availability setup. Its services still share one machine.
Before serving real traffic¶
- Work through the production checklist. Use Production requirements for RPS targets to size resources, concurrency, provider quotas and qualification.
- Keep secrets in environment variables or a secret manager.
- Use protected PostgreSQL and Redis services with backups and monitoring.
- Run database migrations once before starting new application replicas.
- Keep health and metrics endpoints on a trusted network.
- Test an upgrade, rollback, backup, and restore before you need them.
Keep the system healthy¶
- Monitor requests and dependencies
- Upgrade or roll back
- Back up and restore data
- Respond to incidents
- Tune provider connections
- Harden security
Release-specific migration and feature rollout instructions are kept in Operations details. Use one only when the release notes for your target version link to it.