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

# Custom providers — Connect local models and any API endpoint

> Connect local models, self-hosted LLMs, and cloud AI platforms to stagewise. Use OpenAI-compatible, OpenAI Responses, Anthropic, Azure, Bedrock, Vertex, or local endpoints.

Custom endpoints connect stagewise to any model, anywhere — including models running locally on your machine. stagewise supports multiple API styles: OpenAI-compatible chat completions, OpenAI Responses, Anthropic, Google, Azure OpenAI, Amazon Bedrock, and Google Vertex AI.

## Creating a custom endpoint

<Steps>
  <Step title="Open Models & Providers">
    Go to **Settings → Agent → Models & Providers**.
  </Step>

  <Step title="Open custom providers">
    Click **Custom Providers** in the top-right corner of the built-in providers section.
  </Step>

  <Step title="Add a provider">
    Click **Add Provider**.
  </Step>

  <Step title="Configure the provider">
    <Info>
      Unless your provider states otherwise, use **OpenAI (Chat Completions)** — the de facto standard for self-hosted and proxy services.
    </Info>

    Fill in:

    * **Name** — Display name
    * **Provider type** — The API specification your endpoint implements
    * **Base URL** — The endpoint URL
  </Step>
</Steps>

## Supported API specifications

| Spec                      | Example services                                 |
| ------------------------- | ------------------------------------------------ |
| `openai-chat-completions` | Ollama, LM Studio, vLLM, LiteLLM, Together AI    |
| `anthropic`               | Self-hosted Claude, Anthropic-compatible proxies |
| `openai-responses`        | OpenAI Responses API endpoints                   |
| `google`                  | Self-hosted Gemini-compatible services           |
| `azure`                   | Azure OpenAI Service                             |
| `amazon-bedrock`          | AWS Bedrock                                      |
| `google-vertex`           | Google Vertex AI                                 |

## Cloud platform configuration

### Azure OpenAI

Additional fields: **Resource name** and **API version** (e.g., `2024-02-01`).

### Amazon Bedrock

Choose an **Authentication Method**:

| Method                       | What to enter                                                                                                                              |
| ---------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------ |
| **Access Keys**              | **AWS Region**, **Access Key ID**, and **Secret Access Key**                                                                               |
| **Named Profile**            | Optional **AWS Region** override and an **AWS Profile** from `~/.aws/config` or `~/.aws/credentials`                                       |
| **Default Credential Chain** | Optional **AWS Region** override. stagewise resolves credentials from environment variables, shared credentials, ECS/EC2 metadata, and SSO |

SSO profiles require an active session. If requests fail with an expired-token error, run `aws sso login --profile <name>`.

### Google Vertex AI

Additional fields: **Project ID**, **Location** (e.g., `us-central1`), and **Google credentials** (service account JSON).

## Local models via Ollama

A typical setup for running models locally:

1. Run Ollama on your machine: `ollama serve`
2. Pull a model: `ollama pull llama3`
3. Create a custom endpoint in stagewise:
   * **Provider type**: `openai-chat-completions`
   * **Base URL**: `http://localhost:11434/v1`
4. [Add a custom model](/reference/custom-models) with the model ID Ollama expects (e.g., `llama3`)
5. Select the model from the model picker

## Model ID mapping

Some endpoints use different model identifiers. Use **Model ID Mapping** to remap:

```json theme={null}
{
  "claude-sonnet-4-6": "claude-v2"
}
```

This tells stagewise: "when Sonnet 4.6 is selected, request `claude-v2` from this endpoint."

## Using your custom endpoint

After creating the endpoint, go to any built-in provider's configuration and switch to **Custom endpoint** mode. Select your endpoint from the dropdown. All models for that provider route through your custom endpoint.

## Related

* [Models & providers concept →](/core-concepts/models-and-providers)
* [Import your subscriptions →](/reference/byok-setup)
* [Add custom models →](/reference/custom-models)
* [Models & providers configuration →](/reference/models-and-providers)