# Plan a deployment

Source: [https://docs.aventail.co.uk/docs/guides/deployment](https://docs.aventail.co.uk/docs/guides/deployment)

Validate the model, device and application together before a wider rollout.

### Prove a deployment before expanding

Process overview; detailed actions and examples are below.

1.  **Prove a deployment before expanding**
    
    Start with one test device and define success and recovery criteria.
    
    -   Compatibility: Check format, engine and capacity
    -   Representative test: Deploy, serve and inspect results
    -   Controlled rollout: Monitor and preserve a recovery option

**Read the deployment-planning guide**

## Start with a test device

1.  Confirm the model's [compatibility](https://docs.aventail.co.uk/docs/introduction/versions-and-supported-models) and the target device's available capacity.
2.  [Deploy and serve it](https://docs.aventail.co.uk/docs/getting-started/quickstartguide#deploy).
3.  Send representative inputs and check response quality, latency and failure handling.
4.  Record the model, Link version, configuration and observed results.

For organisation devices, use [configuration revisions and rollouts](https://docs.aventail.co.uk/docs/fleet/config-and-rollouts).

## Plan for interruptions

Decide how your application handles an unavailable endpoint, a stopped model or a device update. Test that behaviour before depending on the service. Redundancy and failover require an application and deployment design; they are not implied by registering more devices.

## Monitor the outcome

Watch device reachability, resource use and deployment errors in Control. Measure request latency and failures in your application too.

Define acceptance and rollback criteria for your workload before a wider rollout. Keep the previous working configuration and model available; do not assume stopping a rollout reverses changes already applied.