Skip to main content
Every customer runs their own Azure environment. Operating one means keeping that environment healthy, its secrets fresh, its scheduled jobs running, and its data recoverable.

​
The environment

Infrastructure is Terraform, with per-environment configurations and remote state.

​
Scheduled jobs

Several capabilities depend on jobs firing on time. They are triggered by scheduled workflows calling protected endpoints with a shared CRON_SECRET: The AI-coach learning cycle is the exception: it has no built-in schedule. An operator with system:configure triggers it from the admin button, or you point your own scheduler at its endpoint using CRON_SECRET as a bearer token.
Scheduled workflows run the copy of their file on the default branch, which is develop. A workflow change is live on the schedules as soon as it lands there.

​
Day-to-day

Deployment

The pipeline, environments, and how a release reaches production.

Environment variables

Every setting, what it does, and which belong in Key Vault.

Feature flags

Switching modules off per installation.

Monitoring

Logs, correlation, alerts, and health checks.

Runbooks

Rollback, incidents, restore, secret rotation, customer onboarding.

The docs site

Running and deploying this documentation, including the custom domain.

​
Playbook conventions

The playbooks in docs/operations/ follow two rules that make them usable at 03:00:
  • Every playbook opens with a “When you reach for this” line, so on-call knows in one sentence whether they are in the right place.
  • Every playbook ends with a Verification section. If you cannot tick every box, the procedure is not done.
They are deliberately customer-agnostic and use placeholders (<app>, <rg>, <vault>). Per-customer resource inventories, contacts and decision logs live in a customer overlay under docs/customers/<customer>/.

​
Keeping the playbooks true

Update the relevant playbook in the same PR as any production change that affects operations. A playbook that describes last quarter’s infrastructure is worse than no playbook, because it is trusted.