TokenDocs

English

简体中文

English

简体中文

主题

Last verified:

Verification scope not recorded

WorkBuddy Guide ​

WorkBuddy is Tencent's desktop AI agent workspace. It supports local file workflows, Skills, MCP, automations, and custom models. Its OpenAI-compatible custom model support can connect to TokenFlux.

Download and Install ​

  1. Open the official WorkBuddy download page. You can also start from the China website or the international website.
  2. Choose the build that matches your device:
    • Apple Silicon Mac: choose arm64.
    • Intel Mac: choose x64.
    • Windows: choose the Windows installer.
    • UnionTech UOS / Kylin: follow the official download page's instructions to install from your system's app store.
  3. On macOS, open the downloaded DMG and drag WorkBuddy.app into Applications. On Windows, follow the installer prompts.
  4. On first launch, WorkBuddy prepares its local runtime. Keep the network available and wait for the welcome screen instead of force-quitting the app.
WorkBuddy preparing its runtime during the first launch

When preparation finishes, the welcome screen appears. Click Log in to continue.

WorkBuddy welcome screen with the login button outlined in red

Sign In ​

WorkBuddy opens the Tencent CodeBuddy sign-in page in your browser.

  1. Read the user agreement and privacy policy.
  2. Select the agreement checkbox and click Agree.
  3. Authenticate with WeChat, phone, email, or SSO. Available methods can vary by region and account type.
  4. Return to WorkBuddy after the browser reports a successful sign-in.
Tencent CodeBuddy sign-in page with the agreement checkbox and Agree button outlined in red

Connect TokenFlux ​

1. Prepare an API Key ​

Follow Create API Key, then check the model marketplace for the exact model IDs available to that key's group.

2. Open Custom Model Settings ​

After signing in to WorkBuddy, open:

  1. The avatar or account menu.
  2. System Settings.
  3. Models.
  4. Add Model.
WorkBuddy System Settings window with Models outlined in red in the sidebar

The Models page displays the local configuration path and the Add Model button.

WorkBuddy Models settings page with the Add Model button outlined in red

The Models page displays the actual local configuration path. WorkBuddy Desktop normally uses ~/.workbuddy/models.json; the official model configuration guide also documents migration compatibility with existing ~/.codebuddy/models.json configurations.

3. Enter the Model Configuration ​

Select Custom at the bottom of the provider list. WorkBuddy's custom entry currently requires an OpenAI-compatible API, which TokenFlux provides.

Selecting the red-outlined Custom option from the provider list in WorkBuddy

Then complete the form:

FieldRecommended value
ProviderCustom
Endpointhttps://tokenflux.dev/v1/chat/completions
API KeyTokenFlux API key
Model NameThe exact model ID from the marketplace, such as gpt-6-astra
Tool CallingEnable it; gpt-6-astra supports tool calling
Image InputEnable it so WorkBuddy can send images to the model
ReasoningEnable it to use the reasoning capability of gpt-6-astra
Custom ProtocolLeave disabled to use the OpenAI Chat Completions-compatible protocol
WorkBuddy custom model form configured for TokenFlux and gpt-6-astra with required fields outlined in red

Enter the complete endpoint

Enter the complete endpoint ending in /chat/completions. WorkBuddy 5.3.5 does not automatically append the request path to /v1.

Leave Custom Protocol disabled for TokenFlux

When enabled, WorkBuddy skips standard OpenAI Chat Completions path validation. It is intended for non-standard gateway or proxy routes and is not required for TokenFlux's standard OpenAI-compatible endpoint.

When using a composite key, include the group prefix in the model name, for example GPT/gpt-6-astra. The prefix and model ID must exactly match the composite-key mapping.

4. Save and Verify ​

  1. Click Save. The model is written to the local configuration file shown on WorkBuddy's Models page.

    WorkBuddy custom model form with the Save button outlined in red

  2. After saving, confirm that gpt-6-astra appears under Saved Models.

    The gpt-6-astra card outlined in red in WorkBuddy's Saved Models list

  3. Return to a new task, open the model selector, and choose gpt-6-astra under Custom Models.

    The gpt-6-astra custom model outlined in red in WorkBuddy's model selector

  4. Send a message, such as 你好 ("Hello"), and confirm that the selected model responds.

    WorkBuddy successfully responding with gpt-6-astra, with the model name in the response footer outlined in red

  5. Once basic chat succeeds, test a tool-enabled task, such as asking WorkBuddy to create and read a text file in a temporary directory.

Troubleshooting ​

The saved model does not appear ​

Reopen the model selector or create a new task. Enterprise administrators can disable personal custom models, so enterprise users may need an administrator to enable the policy.

The agent does not call tools ​

Edit the model and enable Tool Calling, then confirm that the upstream model actually supports tool calls. A client-side switch cannot add tool support to an incompatible model.

TokenFlux Documentation

© 2026 TokenFlux