APIMODELSAPIMODELS
WorkBuddy · custom model · OpenAI-compatible

WorkBuddy custom model setup for apimodels.app

WorkBuddy is Tencent's desktop AI agent workspace. Its custom model feature connects model services beyond the built-in ones: you enter a URL, an API key and a model name, and WorkBuddy forwards the requests. With apimodels.app, one key gives you Claude, GPT, DeepSeek and more, billed per token against your apimodels balance. Below: setup steps, MCP and the common errors, with every WorkBuddy-side statement checked against its official docs.

The three fields to fill in

URLhttps://api.apimodels.app/v1/chat/completions
API KeyAn sk_… key from the apimodels console
Model nameclaude-fable-5-2 · gpt-6.2-sol · deepseek-v4.1-flash

No key yet? Create one in Console → Chaves de API. Give WorkBuddy a key of its own; deleting it later cuts access immediately.

Add the custom model in 4 steps

  1. Click your avatar → Settings → Models (设置 → 模型) and add a model.
  2. Choose Custom (自定义) as the provider; any service not in the provider list goes here.
  3. Set URL to https://api.apimodels.app/v1/chat/completions, API Key to your apimodels sk_… key, and the model name to the model id.
  4. Save. Custom models get their own group in the chat model picker; select one and start.
  • Enter the URL all the way to /chat/completions. Leave Custom protocol (自定义协议) under advanced settings off, which is the default: off, WorkBuddy checks the address against the standard /chat/completions path; on, it uses your URL as is. The full URL works either way.
  • The model name must match the id on our model page character for character, dots, dashes and lowercase included: deepseek-v4.1-flash, not DeepSeek-V4.1-Flash.
  • The config, API key included, is stored only on your machine in workbuddy/models.json and is not uploaded; manage it in the UI, no need to edit the file.

Which model id to use

WorkBuddy works through tool calls to read files and operate apps, and all three below support tool calling. For another model, add one more custom model with its id; the full list is in the model catalog. Prices are on each model page.

Claude Fable 5.2claude-fable-5-2Long multi-step tasks, complex documents and code; 1M context.
GPT-6.2 Solgpt-6.2-solA sensible default for everyday tasks; 1.05M context, function calling.
DeepSeek V4.1 Flashdeepseek-v4.1-flashFast, for many simple tasks; 1M context, tool calling.

Add the apimodels MCP server for images and video

MCP (Model Context Protocol) plugs external tools into WorkBuddy. The custom model covers chat; with the apimodels MCP server installed, WorkBuddy can also call generate_image, generate_video, text_to_speech and more, using the image, video and speech models on the site. In the WorkBuddy desktop app open the sidebar Plugins (插件) → MCP servers → Configure MCP and add the block below to mcp.json (user-level file: ~/.workbuddy/mcp.json; for one project only: <project>/.workbuddy/mcp.json):

{
  "mcpServers": {
    "apimodels": {
      "command": "npx",
      "args": ["-y", "apimodels-mcp"],
      "env": {
        "APIMODELS_API_KEY": "sk_…your_apimodels_key…"
      }
    }
  }
}
  • Node.js must be installed (it runs through npx). A green status light means it is connected; on red, check the JSON brackets and quotes, and Node.js.
  • You can also add it under Experts · Skills · Connectors → Connectors → Custom connector (自定义连接器), with the same fields.
  • Then just ask in plain words, e.g. "use apimodels to make a 16:9 product image". Generations bill at each model page's price; failed ones are not charged. Tool list: /docs/mcp.

Does WorkBuddy charge for custom models?

WorkBuddy's docs say all costs of a custom model (tokens, subscriptions and so on) are paid by you to the third party. With apimodels.app, that means our per-million-token price on the model page, deducted from your apimodels balance, and failed requests are not charged. The WorkBuddy Enterprise docs on Tencent Cloud add that web search uses a built-in model to summarize results, which consumes a small amount of WorkBuddy credits.

Common errors

SintomaCausa / correção
404The URL is wrong, e.g. /v1 or /chat/completions appears twice. Use https://api.apimodels.app/v1/chat/completions.
401 / Invalid or missing API keyThe key is wrong, has stray spaces, or was deleted in the console. Copy an sk_… key again.
WorkBuddy error 4100The custom model has no URL or API key, so nothing was sent. Click Go to settings (前往配置) in the message, fill both in and save; no restart needed.
Error about the model nameThe model name differs from our model id. Copy it from the model page, or list every id with GET https://api.apimodels.app/v1/models.
402 / insufficient balanceYour apimodels balance has run out; top up in the console.

Sources (checked 2026-11-18): WorkBuddy docs Model configuration · MCP · Connectors · Error codes · Tencent Cloud WorkBuddy Enterprise model configuration. WorkBuddy is a Tencent product and is not affiliated with apimodels.app.

Perguntas frequentes

What does a custom model mean in WorkBuddy?

It lets WorkBuddy run on a model service you choose instead of its built-in models: you enter a URL, an API key and a model name under Settings → Models, and WorkBuddy only forwards the requests. WorkBuddy's docs state that this config, API key included, is stored only on your machine in workbuddy/models.json and not uploaded, and that you pay the model provider directly. For apimodels.app, the URL is https://api.apimodels.app/v1/chat/completions and the model name is our model id, such as claude-fable-5-2.

How do I configure a custom model in WorkBuddy?

Click your avatar → Settings → Models, add a model and choose Custom. Fill in three fields: URL https://api.apimodels.app/v1/chat/completions, API Key an sk_… key from the apimodels console, and the model name as the exact model id, e.g. gpt-6.2-sol, claude-fable-5-2 or deepseek-v4.1-flash. Save, then pick it from the custom group in the chat model picker. Leave the Custom protocol switch under advanced settings off, its default, so WorkBuddy checks the address against the standard /chat/completions path.

Why does my WorkBuddy custom model show an error?

Match the error: 404 means the URL is wrong, for example /v1 or /chat/completions written twice; it should be https://api.apimodels.app/v1/chat/completions. 401 Invalid or missing API key means the key is wrong or was deleted; create a new one in the console. WorkBuddy error 4100 means the custom model has no URL or API key, so click Go to settings and fill them in. An error about the model name means it differs from our model id; copy the id from the model page. 402 means the apimodels balance has run out.

Does WorkBuddy still charge for a custom model?

WorkBuddy's docs say all costs of a custom model, tokens or subscriptions, are paid by you to the third party. With apimodels.app that is our per-million-token price on each model page, deducted from your apimodels balance, and failed requests are not charged. The WorkBuddy Enterprise docs on Tencent Cloud add that web search uses a built-in model to summarize results, which consumes a small amount of WorkBuddy credits.

What is MCP in WorkBuddy, and how do I add the apimodels MCP server?

MCP (Model Context Protocol) plugs external tools into WorkBuddy. In the desktop app open Plugins → MCP servers → Configure MCP and add to mcp.json (user level: ~/.workbuddy/mcp.json): {"mcpServers": {"apimodels": {"command": "npx", "args": ["-y", "apimodels-mcp"], "env": {"APIMODELS_API_KEY": "sk_…"}}}}. Node.js must be installed. Once the status light is green, ask in plain words for images, video or voice-over; generations bill at each model page's price. The same server can be added as a custom connector under Experts · Skills · Connectors.

Where do I get an API key for WorkBuddy, and is it safe to enter?

The API key in a WorkBuddy custom model comes from the model service you connect. For apimodels.app, create an sk_… key under Console → API Keys. WorkBuddy's docs state the key is saved only locally in workbuddy/models.json and never uploaded. Give WorkBuddy its own key; deleting it in the console cuts access immediately.