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

# Custom apps

> Connect your app to Magnitude's local inference API.

Connect your own app to Magnitude to run inference with downloaded models. Apps that support a custom OpenAI-compatible base URL can connect without a Magnitude-specific integration.

## Before you start

1. Install and open Magnitude.
2. Download a model in **Discover**.
3. Keep Magnitude running while your app uses it.

Magnitude loads the requested model on demand. The first response can take longer while it loads. To load it ahead of time, click **Load model** in **My Models**.

## Configure an OpenAI-compatible app

| Setting  | Value                                                   |
| -------- | ------------------------------------------------------- |
| Provider | OpenAI-compatible                                       |
| Base URL | `http://127.0.0.1:10100/inference/v1`                   |
| API key  | `magnitude-local` if your app requires a value          |
| Model    | An exact model ID returned by the models endpoint below |

Local inference does not require an API key. `magnitude-local` is a placeholder for clients that require a nonempty key.

Use the base URL above when your app appends `/chat/completions` or `/responses` itself. If it asks for a complete endpoint URL, use the corresponding full path from the endpoint table below.

## Find a model ID

List the models available for inference:

```sh theme={"theme":{"light":"github-light","dark":"github-dark"}}
curl http://127.0.0.1:10100/inference/v1/models
```

The response is an OpenAI-compatible model list with a `data` array. Copy the `id` of the model you want to use. Pass that exact value as `model` in requests, rather than the model's display name.

## Send a request

Replace `MODEL_ID` with an ID from the models response:

```sh theme={"theme":{"light":"github-light","dark":"github-dark"}}
curl http://127.0.0.1:10100/inference/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "MODEL_ID",
    "messages": [
      {"role": "user", "content": "Explain prompt caching in one paragraph."}
    ],
    "max_tokens": 256
  }'
```

Read the generated text from `choices[0].message.content`. To stream the response, add `"stream": true` to the request body and use `curl -N` to display events as they arrive. Streaming responses use server-sent events (SSE).

These shell examples use macOS/Linux quoting. On Windows, use `curl.exe` with quoting appropriate for your shell, or send the same JSON body from your application.

## HTTP endpoints

All paths below are relative to `http://127.0.0.1:10100`.

| Method | Path                                            | Purpose                                                            |
| ------ | ----------------------------------------------- | ------------------------------------------------------------------ |
| `GET`  | `/health`                                       | Check whether the service is ready; returns `200` when ready.      |
| `GET`  | `/inference/v1/models`                          | List models available for inference.                               |
| `POST` | `/inference/v1/chat/completions`                | OpenAI-compatible Chat Completions, with JSON or streaming output. |
| `POST` | `/inference/v1/responses`                       | OpenAI-compatible Responses, with JSON or streaming output.        |
| `POST` | `/inference/anthropic/v1/messages`              | Anthropic-compatible Messages.                                     |
| `POST` | `/inference/anthropic/v1/messages/count_tokens` | Count input tokens for an Anthropic-format request.                |

For an Anthropic-compatible client that appends `/v1/messages`, use `http://127.0.0.1:10100/inference/anthropic` as its base URL. Messages requests require `max_tokens`.

Supported inputs and features depend on the selected model. These endpoints provide local model inference; your application handles conversation history and executes any tools requested by the model.

## Local access and troubleshooting

The API listens on the same machine as Magnitude. `127.0.0.1` in a container or on another machine refers to that environment, not the machine running Magnitude. Direct LAN or internet access is not provided. Browser access is restricted to local origins; a page hosted on an external website cannot call the API directly through ordinary cross-origin requests.

* **Connection refused:** open Magnitude, or run `magnitude service start` using the bundled CLI.
* **`/health` works but inference returns `404`:** check the URL prefix. Use `/inference/v1/chat/completions`, not `/v1/chat/completions` at the server root.
* **Model unavailable:** download a model and copy its exact ID from `/inference/v1/models`.
* **Slow first response:** allow time for model loading, or load the model ahead of time in **My Models**.

See [Inference](/inference) for loading, memory, and performance behavior, and the [CLI reference](/reference) for model and service management.