# Custom OpenAI-compatible endpoints

> Connect any public OpenAI-compatible server, such as vLLM, a hosted inference service or your own proxy, and call its models through DigitSpace.

Source: https://digitspace.ai/docs/keys/custom · Last checked: 2026-10-31

Any server that speaks the OpenAI Chat Completions API at a public HTTPS address can be added as a **Custom (OpenAI-compatible)** provider.

## Steps

1. In **Provider keys → Add a provider key**, choose **Custom (OpenAI-compatible)**.
2. Enter a **Name** in lowercase words joined by hyphens, such as `my-vllm`. It becomes part of the model IDs.
3. Enter the **Base URL**, the address that ends where `/chat/completions` and `/models` begin, such as `https://llm.example.com/v1`. It must be a public HTTPS address.
4. Paste the server's API key, and save.

DigitSpace reads the server's `/models` list with the key and saves those models.

## Calling its models

Use the name you gave the key in front of the model:

```json
{ "model": "my-vllm/llama-3.3-70b", "messages": [ … ] }
```

`custom/llama-3.3-70b` matches the model on any of your custom keys. If no catalogue model has the same ID, the bare model ID works too. Use **Sync models** in the key's menu after adding models to the server.

## Things to know

- These requests run on your server, so they're never charged credits. Their cost shows as $0 unless you set prices for the models (Enterprise, under **Models**).
- Routing treats these models as able to do anything a request asks (images, tools, JSON schema, reasoning). Make sure your server can.
- They aren't available on a [DevPass](/docs/devpass).