Self-Hosted Deployment
Run Align in your own infrastructure. Your decision graph lives in your own database.
Choose your install
| Way to run Align | Status | Start here |
|---|---|---|
| Align cloud | Fastest. We run it. | Quickstart |
| Helm on your own cluster | Supported. | Installation |
| One-VM pilot | Pilot preview, by arrangement. Ask us first. | Pilot |
| Offline bundle | On request. We build it per release. | Offline install |
Growth and Enterprise plans can both self-host. Licenses are covered in Licensing.
What you need
| Need | Detail |
|---|---|
| Image access | The 12-digit AWS account ID we grant pull access to, or the offline bundle |
| Kubernetes and Helm | Kubernetes 1.25+, Helm 3.10+ (the minimum versions) |
| PostgreSQL with pgvector | 15 or newer. The chart can run a bundled Postgres 16 for trials. |
| An LLM key | OpenAI or Anthropic, or your own model server |
| Sign-in | An OIDC app in your identity provider |
| Connectors | An OAuth app for each tool you turn on. See OAuth apps |
| A license | Align sends the license and its public key at onboarding |
| Ingress, two DNS names and TLS | You provide these on Helm |
| Redis | Needed above one gateway pod |
Sizing is in Infrastructure.
What runs
| Component | Purpose | Required? |
|---|---|---|
| UI | Web app | Yes |
| Gateway | API server | Yes |
| Brain | AI service | Yes |
| Connectors | One pod per tool. Six are on by default. | Yes |
| PostgreSQL | Database | Yes |
| Redis | Job state and shared session revocation | Above one gateway pod |
| vLLM server | Your own GPU model | No |
Redis and gateway pods
The gateway runs one pod by default. With more than one pod, Redis is required. Without it, a revoked session stays valid on the pods that did not handle the revocation. The chart refuses to render that setup unless you set gateway.redis.url or gateway.redis.secretName.
gateway:
replicaCount: 2
redis:
url: "redis://your-redis:6379"
With one pod and no Redis, job state is lost on restart. Any Redis-compatible server works.
The job queue is in memory unless you set SQS queue URLs. Set them if queued import work must survive a pod restart.
Quick start
The chart is an OCI artifact. It is not in a Helm repository. Installation covers the registry login and the secrets you create first.
helm install align \
oci://$ALIGN_REGISTRY/align/charts/align \
--version "$ALIGN_CHART_VERSION" \
--namespace align \
--create-namespace \
--values values.yaml
For production, copy values-production.template.yaml from the chart and fill it in.
Documentation
| Guide | Description |
|---|---|
| Pilot | The one-VM install, in preview |
| Requirements | What to have ready |
| Infrastructure | What to provision, and sizing |
| Installation | Step-by-step Helm install |
| Offline install | The offline bundle |
| Verify images | Check signatures before you deploy |
| Configuration | Helm values |
| Values reference | Every value, grouped |
| Secrets | Secrets management |
| OAuth apps | Register your own OAuth apps |
| LLM setup | Providers and embeddings |
| Network | Proxies and outbound domains |
| Observability | Metrics and traces |
| Licensing | License keys and seats |
| Key rotation | Rotate encryption keys |
| RLS cutover | The restricted database role |
| Upgrades | Upgrade and migration guide |
Support
- Email: support@align.tech
- Documentation: docs.align.tech