# WorkBuddy

> Connect WorkBuddy to the LMU AI relay via a Custom OpenAI-compatible model — set the Base URL and sk- key to call Claude, GPT, and Chinese models.

URL: https://docs.lmuai.ai/docs/tools/workbuddy





WorkBuddy is Tencent's AI coding assistant (same family as CodeBuddy). It supports connecting to any OpenAI-compatible API through a **Custom (OpenAI-compatible) model**, so you can point it at the LMU AI relay and use a single `sk-` key to call Claude, GPT, and Chinese models.

***

## Before you start [#before-you-start]

* An LMU AI API key that starts with `sk-`: register and generate one in the [LMU AI console](https://api.lmuai.ai).

* Note the LMU AI OpenAI-compatible address (Base URL), **with** `/v1`:

  ```
  https://api.lmuai.ai/v1
  ```

* Make sure your key's group has the upstream for the model you want to use (available models depend on the group).

<Callout type="warn" title="Do not use a Claude Max group key">
  The Claude Max group supports only the Anthropic protocol and **cannot** be filled into an OpenAI-protocol address such as `https://api.lmuai.ai/v1`, so it will not work with WorkBuddy's custom model. Use a **pay-as-you-go** or **standard subscription** group key instead.
</Callout>

***

## Method 1: Add a custom model in the UI (recommended) [#method-1-add-a-custom-model-in-the-ui-recommended]

1. Open WorkBuddy and go to the model settings: either via **Settings → Model**, or open the model picker in the chat box, scroll to the bottom, and choose **Configure custom models**.

2. Click **Add Model** and set the provider / type to **Custom**.

3. Fill in these three fields:

   | Field               | Value                                                       |
   | ------------------- | ----------------------------------------------------------- |
   | Endpoint / Base URL | `https://api.lmuai.ai/v1`                                   |
   | API Key             | the `sk-` key generated in the console                      |
   | Model Name          | an LMU AI model ID, e.g. `gpt-5.6-sol` or `claude-sonnet-5` |

4. (Optional) Toggle Tool Call, Image input, etc. according to the model's capabilities.

5. Save, return to the model picker, select the model you just added, and start chatting.

<img alt="The WorkBuddy &#x22;Add Model&#x22; dialog filled in: provider set to Custom, endpoint stopping at https://api.lmuai.ai/v1" src="__img0" />

<Callout type="info" title="Do not write /chat/completions into the endpoint twice">
  WorkBuddy automatically appends `/chat/completions` to the endpoint. So stop the endpoint at `https://api.lmuai.ai/v1` and do not add `/chat/completions` — otherwise it becomes `…/v1/chat/completions/chat/completions` and returns a 404.
</Callout>

***

## Method 2: Edit models.json (precise, version-controllable) [#method-2-edit-modelsjson-precise-version-controllable]

WorkBuddy / CodeBuddy also stores custom models in a config file, which suits bulk or team-wide setup.

The config file lives in a `.codebuddy` folder in your home directory (some versions use `.workbuddy` — use whichever actually exists on your machine):

* **Windows (user level)**: `C:\Users\<your-username>\.codebuddy\models.json`
* **macOS / Linux (user level)**: `~/.codebuddy/models.json`
* **Project level**: `<your-project-dir>\.codebuddy\models.json`

On first use, install, sign in, and open any project once so the config directory is created. Example (replace `apiKey` with your own `sk-` key):

```json
{
  "models": [
    {
      "id": "gpt-5.6-sol",
      "name": "LMU AI GPT-5.6 Sol",
      "vendor": "LMU AI",
      "url": "https://api.lmuai.ai/v1/chat/completions",
      "apiKey": "sk-your-key",
      "maxInputTokens": 128000,
      "maxOutputTokens": 8192,
      "supportsToolCall": true,
      "supportsImages": false
    },
    {
      "id": "claude-sonnet-5",
      "name": "LMU AI Claude Sonnet 5",
      "vendor": "LMU AI",
      "url": "https://api.lmuai.ai/v1/chat/completions",
      "apiKey": "sk-your-key",
      "maxInputTokens": 128000,
      "maxOutputTokens": 8192,
      "supportsToolCall": true,
      "supportsImages": false
    }
  ],
  "availableModels": ["gpt-5.6-sol", "claude-sonnet-5"]
}
```

Notes:

* Here `url` must be the **full path** `https://api.lmuai.ai/v1/chat/completions` (unlike Method 1, it does include `/chat/completions`).
* `maxInputTokens`, `maxOutputTokens`, `supportsToolCall`, and `supportsImages` are **client-side declared values** — adjust them to the actual capabilities of the model you chose.
* Save the file as **UTF-8 without BOM**; a BOM may cause it to fail to load.
* You can also set the key as an environment variable and reference it, e.g. `"apiKey": "${LMU_API_KEY}"` (`setx LMU_API_KEY "sk-..."` on Windows, `export LMU_API_KEY=sk-...` on macOS / Linux).
* After editing, **fully quit and restart** WorkBuddy, then select the model in the picker.

***

## Available models [#available-models]

Common model IDs (the ones you can actually use depend on your group; see the full list on [Models](../guide/models)):

* **GPT**: `gpt-5.6-sol`, `gpt-5.6-terra`, `gpt-5.6-luna`, `gpt-5.5`
* **Claude**: `claude-opus-5`, `claude-sonnet-5`, `claude-haiku-4-5`
* **Chinese**: `deepseek-v4-pro`, `qwen3.8-max-preview`, `glm-5.2`, `kimi-k3`

WorkBuddy uses the OpenAI-compatible protocol, and the LMU AI backend performs protocol conversion automatically, so Claude / Chinese model IDs work too — provided your group has the matching upstream.

***

## FAQ [#faq]

### 401 / 403 error [#401--403-error]

Check in order:

1. Whether the API Key is an `sk-` key generated in the LMU AI console (not an official OpenAI / Anthropic key).
2. Whether the endpoint is correct: `https://api.lmuai.ai/v1` (with `/v1`).
3. Whether your key's group includes the model you are calling.

### 404 / Not Found [#404--not-found]

Usually the endpoint has `/chat/completions` written twice, or is missing `/v1`:

* Method 1 (UI): stop the endpoint at `https://api.lmuai.ai/v1`; WorkBuddy appends `/chat/completions`.
* Method 2 (models.json): set `url` to the full `https://api.lmuai.ai/v1/chat/completions`.

### "Model unavailable" / "No available accounts" [#model-unavailable--no-available-accounts]

You called a model outside your group's range (for example, using a Claude Max group to call GPT). Check the models your current group can call under "Available models" in the console, or switch groups. See [API protocols](../guide/api-protocols).

### Changes have no effect [#changes-have-no-effect]

After editing `models.json`, **fully quit and restart** WorkBuddy; save the file as **UTF-8 without BOM**. For a model added through the UI, just reselect it in the model picker after saving.

***

## Notes [#notes]

* Enter only the LMU AI relay address as the endpoint; do not enter the official OpenAI / Anthropic address.
* The same `sk-` key can be used with other tools at the same time; WorkBuddy uses the OpenAI-compatible protocol.
* The LMU AI overseas API domain connects directly from overseas, with no proxy needed.

For more, see [API protocols](../guide/api-protocols) and the [FAQ](../guide/faq).