Production
Pick a production setup for self-hosted Future AGI, then harden it and keep it running: secrets, TLS, backups, monitoring, upgrades and rollback.
In this page
Before real traffic reaches the instance, pick the setup you run it on, work through the two go-live pages below, then keep the rest as runbooks.
Run production on Distributed with the production overlay or on Helm (Kubernetes). Standalone suits a single host and a small team: the same checklist applies, but it has no production overlay.
Choose a production setup
- Distributed with the production overlay: one container per service on one host. The overlay in
deploy/refuses to start until its secrets and image versions are set. See Start Distributed with the production overlay. - Helm (Kubernetes): one Deployment per service, with external datastores by default. See Install for production.
- Standalone: three containers on one host, for a small team. Work through the same checklist: Standalone has no production overlay, so nothing stops it from starting on weak settings.
Choose before you add data. There is no supported way to move a Standalone install’s data to Distributed or Helm later: see Switch setups.
Warning
A default install is not ready to expose. ./bin/install generates secrets on a fresh install, but not the two passwords of the observed-attribute catalog’s ClickHouse users. Installs made before Standalone became the default, and Compose installs started without the installer, may still run on the defaults published in the repository, with no ClickHouse password. Nothing terminates TLS by default: once people reach the instance on a public host name, the setup screen’s SSL/TLS certificate check fails and blocks a Production launch until you add it.
Before you go live
Checklist
Real secrets, production mode, public URLs, pinned images, and managed data stores
Security & TLS
Terminate TLS in front of the stack, close the plaintext ports, and isolate code evals
Operating it
Backups & restore
Back up and restore Postgres, ClickHouse, object storage, and .env
Monitoring
Watch the health checks, endpoints, and metrics the stack exposes
Upgrades & rollback
Pull a release, roll back safely, and upgrade installs made before Standalone became the default
Dive deeper
Questions & Discussion