Docs › Setup guides · 中文

Use SteadyRouter with WorkBuddy

WorkBuddy, Tencent's desktop AI assistant, can use your own model provider. It talks to custom providers in the OpenAI Chat Completions format, so you add each SteadyRouter model as a Custom model with SteadyRouter's chat endpoint.

Before you start

1. Create a SteadyRouter account.
2. In the console, open API Keys and copy your key. New accounts come with one called Default Key, on Lowest price (see below). To make another, click Create API Key. Keep Smart routing on: if a group fails, the request is retried in the next group within the same call. A new key starts on Auto, or on Lowest price if your balance is under US$1.
3. Install WorkBuddy from its official website and sign in. The desktop app runs on macOS 12 or later (Apple silicon and Intel) and Windows 10 or later (x64).

Add a model

1. Open the model settings and add a new model. In WorkBuddy AI, the international edition, click your avatar → System Settings → Models; in the China edition, open Settings → Model and click Add model. You can also get there from the model selector at the bottom of a conversation.
2. Provider: Custom.
3. URL: the full chat endpoint, including /v1/chat/completions:
   URLCopy

   https://api.steadyrouter.com/v1/chat/completions

   Leave Custom Protocol in the advanced settings off. SteadyRouter uses the standard path, so WorkBuddy doesn't need it.

4. API Key: your SteadyRouter key.
5. Model name: a model name from Model Square, exactly as shown, for example gpt-6.1-sol. Model names have no vendor prefix.
6. Under the advanced settings, tick only the capabilities the model has. The table below lists them for the models we recommend; on Model Square, the Tools, Vision and Reasoning tags on each model card mean the same thing. If you're not sure, leave the box unticked: ticking a capability the model doesn't have causes errors.
7. If the form has a Test Connection button, click it to check the URL and key. Then click Save. Add each model you want to use the same way; they all use the same URL and key.
8. Pick the model in the model selector and send a short message to check the connection.

Recommended models

Each of these is offered by groups from at least two independent suppliers, so smart routing can move a failed request to another group. Tick the boxes as shown.

OpenAI, Google and xAI

Chinese developers

Claude models such as claude-sonnet-5-5 can be added the same way, and they handle tool tasks such as creating and editing files. But WorkBuddy calls them through the OpenAI format: prompt caching and extended thinking are only available through the native Anthropic endpoint, which WorkBuddy doesn't support, so every step pays full price for WorkBuddy's long prompt again. Expect them to cost much more than the models above. Leave Reasoning unticked for Claude models.

Adding several models at once

WorkBuddy keeps custom models in a file called models.json. WorkBuddy AI's docs put it at ~/.workbuddy-ai/models.json (on Windows, %USERPROFILE%\.workbuddy-ai\models.json); the China edition uses ~/.workbuddy/models.json (on Windows, C:\Users\<you>\.workbuddy\models.json). If your settings show a different location, go by that. Instead of adding models one by one, you can add entries to that file. It holds a list ([] in a fresh install); WorkBuddy picks up the change as soon as you save, without a restart. Each entry uses the same values as the form above:

[
  {
    "id": "gpt-6.1-sol",
    "name": "SteadyRouter gpt-6.1-sol",
    "vendor": "SteadyRouter",
    "url": "https://api.steadyrouter.com/v1/chat/completions",
    "apiKey": "sk-your-steadyrouter-key",
    "maxInputTokens": 128000,
    "maxOutputTokens": 8192,
    "supportsToolCall": true,
    "supportsImages": true,
    "supportsReasoning": true
  },
  {
    "id": "deepseek-v4-pro",
    "name": "SteadyRouter deepseek-v4-pro",
    "vendor": "SteadyRouter",
    "url": "https://api.steadyrouter.com/v1/chat/completions",
    "apiKey": "sk-your-steadyrouter-key",
    "maxInputTokens": 128000,
    "maxOutputTokens": 8192,
    "supportsToolCall": true,
    "supportsImages": false,
    "supportsReasoning": true
  }
]

If the file already has other entries, add yours to the existing list rather than replacing the file. The key is stored in plain text on your computer, and WorkBuddy also copies it into its own cache. Don't share or commit this file. If you stop using the key or think it may have leaked, delete it in the SteadyRouter console; removing the model from WorkBuddy is not enough.

Trying it on a small balance

New accounts currently start with US$0.05 of credit. Before each request, the gateway reserves balance based on the input plus the maximum output the app asks for, and an agent like WorkBuddy sends long prompts and runs several steps per task. With a small balance:

1. Use Lowest price under Smart routing on your API key. Default Key is already on it; for another key, choose it when you create the key or later by editing it.
2. Start with a smaller model such as gemini-3.8-flash or qwen3.8-flash. On Lowest price, a request to these costs a small fraction of the same request to larger models such as gpt-6-astra or gemini-4-argon.
3. If the model form offers an output size, choose the smallest, 8K. This lowers the amount reserved per request.

Once you've confirmed everything works, add funds, and switch the key back to Auto if you prefer.

Lowest price often puts Pool groups first: usually cheaper, sometimes less stable. In every smart routing mode, Pool and Mixed groups can be used, so for stability-critical work, turn smart routing off and list only cloud or API groups on the key. See Models and groups.

If something goes wrong

Every error message ends with a request id. Include it when you email support@steadyrouter.com.

More: API docs · Model Square