# Manage models

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

Choose a shared model or upload a custom artifact, then test it on one device.

### Choose and upload models

Control model screens captured 28 September 2026.

1.  **Browse the library**
    
    Your Models contains your uploads; Globally Shared contains shared models. Check compatibility before deploying.
    
    [Screenshot: Shared model library — close-up from Control model library.](https://docs.aventail.co.uk/img/guides/context/models.png)
    
2.  **Upload an artifact**
    
    Enter its name, file, model type, framework and version. Check the current form’s limits before uploading.
    
    [Screenshot: Upload Model form with name, file, model type, framework and version fields.](https://docs.aventail.co.uk/img/control/reference/upload-model.jpg)
    
3.  **Validate on one device**
    
    Deploy to a test device, start serving and check a response before wider distribution. An accepted upload does not prove compatibility.
    
    [Screenshot: Deploy Model dialog for Gemma 4 (X-Small), with Target Device selection and Deploy.](https://docs.aventail.co.uk/img/control/onboarding/deploy-model.png)
    

**Read the model-management guide**

## Browse the library

Open **Models** in Control. **Your Models** contains your uploads; **Globally Shared** contains models made available for use across accounts. Check the task, framework, size and deployment state before choosing one.

## Choose a compatible model

Start with a shared model from the [tested walkthrough](https://docs.aventail.co.uk/docs/getting-started/quickstartguide). Check [compatibility](https://docs.aventail.co.uk/docs/introduction/versions-and-supported-models) before choosing another format or architecture.

## Upload a custom model

Select **Upload Model**, enter a name, choose the file, model type, framework and version, then upload. The form observed on 28 September 2026 listed `.onnx`, `.pb`, `.tflite`, `.pt`, `.pth`, `.gguf`, `.bin` and `.h5`, **up to 500 MB**. Check the current form before preparing a larger file; globally shared models can exceed this upload limit.

An accepted upload is only a stored artifact; validate engine compatibility and required input separately.

After upload, [deploy to one test device](https://docs.aventail.co.uk/docs/getting-started/initial-setup#deploy-a-model), inspect the result, and only then distribute it more widely.

## Capacity and quality

Check disk space, memory, response quality and request latency on the target hardware. Quantisation and context length affect these differently; there is no single minimum-memory figure for every model of a given size.

## Remove a model

-   [Uninstall from one device](https://docs.aventail.co.uk/docs/model_management/removing-models#from-a-device) to free that device's disk space.
-   [Delete from your library](https://docs.aventail.co.uk/docs/model_management/removing-models#from-the-library) only when the artifact is no longer needed.

Keep a recoverable copy of custom model files before deletion. For wider deployments, [plan and test the rollout](https://docs.aventail.co.uk/docs/guides/deployment).