> ## Documentation Index
> Fetch the complete documentation index at: https://docs.uniac.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Deployments

> Versioned snapshots of a Service.

A **deployment** is a single versioned snapshot of a Service. Each `uniac deploy` creates a new deployment with an incremented version number; older versions remain in the registry.

The platform's source of truth for "which version is live" is the deployment's **desired replica count**:

* `NULL` — never promoted; image is in the registry but no instances are scheduled.
* `0` — drained; was live, now scaled to zero.
* `N ≥ 1` — active; the reconciler converges running instances to N.

When you deploy a new version, the platform sets the new version's replica count and zeros out the previous one. The `endpoint-id` for the Service stays stable across all versions, so URLs don't change.

## Rolling back

There is no `uniac` command for rollback. To revert to a prior version, use the dashboard or the [Platform API](/platform-api/overview) to change which version's `desired_replica_count` is non-zero.

## Scaling

Scaling is manual. POST to `/api/services/{service_id}/scale` with `{"target_replica_count": N}`. The platform enqueues a task and the reconciler converges the running instances.

There is no auto-scaling and no per-tenant ceiling beyond a system-wide sanity cap of 100.