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

# Model Distillation

> Train specialized models from real application traffic, compare them with the models you use today, and deploy the best result.

Model Distillation helps product teams turn successful AI interactions into smaller, faster, and less expensive models built for their product. It brings the complete improvement workflow into one place, from collecting examples to deploying a better model.

<CardGroup cols={2}>
  <Card title="Get started" href="/model-distillation/quickstart" arrow="true">
    Connect your application and follow the complete path from real usage to a deployed model.
  </Card>

  <Card title="What Model Distillation provides" href="/model-distillation/overview" arrow="true">
    See how the proxy, Studio, training, evaluations, and routing work together.
  </Card>

  <Card title="API reference" href="/model-distillation/reference/overview" arrow="true">
    Integrate with the management API or call the OpenAI-compatible inference proxy.
  </Card>
</CardGroup>

## The basic idea

Connect an existing AI feature once. Model Distillation collects examples from real usage and lets you turn the best examples into training data.

Train several models, compare their quality with the model you use today, and deploy the best one without rewriting your application.

<Tip>
  Automation can run the entire improvement loop and deploy a winner only when it meets the quality bar you choose.
</Tip>

<Accordion title="API: Agent documentation">
  Start with [`llms.txt`](https://docs.coreweave.com/llms.txt) for a compact map of the customer journey and machine-readable resources. For implementation, give the agent:

  * the [Quick Start](/model-distillation/quickstart) for the intended workflow;
  * the [Management OpenAPI specification](/openapi/model-distillation/management.openapi.yaml) for Studio operations;
  * the [Inference proxy OpenAPI specification](/openapi/model-distillation/proxy.openapi.yaml) for application requests.

  Ask the agent to use only documented public APIs and to show you any routing or compute-consuming request before it sends it.
</Accordion>