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

# Management API

> Programmatically upload, version, and deploy custom model weights with Dedicated Inference.

<Callout icon="lock" color="#b2b1b1ff" iconType="regular">
  This feature is in [Private Preview](/support/preview-releases). For access or more information, [contact us](https://www.cerebras.ai/contact) or reach out to your account representative.
</Callout>

Upload fine-tuned models from your S3 bucket, track multiple versions, and create deployments behind your [dedicated endpoint](/dedicated/overview).

For detailed endpoint documentation, see the [Management API Reference](/api-reference/customer_management_api/list-model-architectures).

## Key Concepts

* **Model Architecture**: The serving compatibility class for a model, such as `gpt-oss-120b` or `zai-glm-4.7`. Each endpoint supports a specific architecture, and custom weights must be compatible with that architecture.
* **Model Version**: An immutable snapshot of model weights. Each upload creates a new version identified by an auto-incrementing integer, such as `1`, `2`, or `3`. Versions can also have aliases like `production` or `v1-stable`.
* **Deployment**: A running model version and its serving configuration. A deployment belongs to an endpoint and can be replaced without changing the endpoint ID used by applications.
* **Endpoint**: A stable inference target identified by a unique endpoint ID, such as `my-org-gpt-oss-120b`. The endpoint routes requests to one or more deployments.
* **Replica**: One running copy of a deployment. Replica count determines the capacity available to the deployment.

<Note>
  The inference API uses the OpenAI-compatible `model` field as a target selector. For Dedicated Inference, pass the endpoint ID in this field. In the Management API, `model` refers to a model version resource name where noted.
</Note>

## Typical Workflow

1. **Upload model weights** — Push your fine-tuned weights from S3 to Cerebras. The upload is asynchronous; you'll receive a version ID to track progress.

2. **Check upload status** — Poll the version status until the sync completes.

3. **Create a deployment**: Once the upload is complete, deploy the version behind your endpoint.

4. **Run inference**: Make requests to your endpoint using the endpoint ID as the `model` field.

5. **Iterate**: Upload new versions as you fine-tune, assign aliases like `production` to track releases, and replace deployments without changing the endpoint ID.

## Authentication

The Management API uses a separate API key from the standard inference API. You can find your Management API key under **Management API keys** on the API keys page in the [Cerebras Cloud console](https://cloud.cerebras.ai).

## S3 Bucket Setup

Before uploading model weights, you need an S3 bucket configured with cross-account access to Cerebras. Your Cerebras representative will provide specific instructions, but the bucket policy must grant cross-account access:

```json theme={null}
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": {
        "AWS": "<cerebras-provided-iam-role-arn>"
      },
      "Action": [
        "s3:GetObject",
        "s3:GetObjectVersion",
        "s3:ListBucket",
        "s3:GetBucketLocation"
      ],
      "Resource": [
        "arn:aws:s3:::<your-bucket-name>",
        "arn:aws:s3:::<your-bucket-name>/*"
      ]
    }
  ]
}
```

Replace `<your-bucket-name>` with your S3 bucket name. The IAM role ARN (`<cerebras-provided-iam-role-arn>`) will be provided by Cerebras and enables secure cross-account access to your model weights.