ZCode is an AI coding assistant from Zhipu AI that supports custom providers compatible with the OpenAI and Anthropic protocols. For HaiToken, the OpenAI path is usually the most stable and broadly compatible choice; if you explicitly want the Claude / Anthropic ecosystem, you can also use the Anthropic path.
What you need
- A HaiToken API key
- HaiToken OpenAI-compatible address:
https://api.haitoken.ai/v1
- HaiToken Claude / Anthropic-compatible address:
https://api.haitoken.ai
- At least one available model ID, such as
deepseek-v4-flash
ZCode is configured entirely through the GUI, so the steps are essentially the same on Windows, macOS, and Linux. If menu placement or dialog layout differs slightly on your system, follow the actual UI — the API key, Base URL, and model ID values remain the same.
Setup Steps
Choose the protocol first
In ZCode, the two common choices are:
OpenAI: recommended first, broader compatibility, and the main flow in this guideAnthropic: also common, useful when you explicitly want the Claude / Messages ecosystem
If you do not have a strong reason, start with OpenAI.
Step 1: Open Model Settings
In the ZCode main interface, click Model Settings in the left sidebar to open the model provider management page.
Step 2: Add a Custom Provider
In the Custom Providers area of the model settings page, click the + Add Provider button to open the provider configuration form.
Step 3: Fill in HaiToken Configuration
In the provider configuration form, enter the following:
| Field | Value |
|---|
| Provider name | haitoken (customizable) |
| Base URL | Use https://api.haitoken.ai/v1 for OpenAI; use https://api.haitoken.ai for Anthropic |
| API Key | Your HaiToken API key |
| API format | Prefer OpenAI; choose Anthropic Messages (/v1/messages) if you explicitly want the Claude ecosystem |
If selecting OpenAI as the API format causes a /responses suffix to appear below the Base URL field, the current configuration uses the Responses API path. HaiToken is compatible with the Chat Completions path (/v1/chat/completions). Select an OpenAI option that does not include /responses to ensure the correct request path.
If you choose the Anthropic path, do not enter https://api.haitoken.ai/v1. This path should use the Anthropic-style gateway prefix https://api.haitoken.ai, and the client should call /v1/messages through that gateway.
Step 4: Add Models
In the Model List area, click + Add Model and enter the HaiToken model ID you want to use, such as deepseek-v4-flash. You can add multiple models.
Step 5: Save the Provider
After filling in the fields, click the Add Provider button at the bottom of the form to save the configuration.
Step 6: Verify Connectivity
After saving, switch to the newly configured HaiToken model in the ZCode chat or coding panel, and send a simple message (such as hello) to verify connectivity.
Notes
| Item | Details |
|---|
| Recommended protocol | OpenAI Chat Completions |
| Common request path | POST /v1/chat/completions |
| Claude-compatible path | POST /v1/messages |
| Auth header | Authorization: Bearer YOUR_API_KEY |
| Model switching | Use the model picker in the ZCode chat or coding panel |
FAQ
| Question | Solution |
|---|
| Model list loads successfully but invocation fails | Check whether the API format is using the Responses API (URL suffix shows /responses). HaiToken is compatible with Chat Completions — select an OpenAI option that does not include /responses. |
| Endpoint is reported as invalid | Make sure the Base URL is https://api.haitoken.ai/v1 without appending /chat/completions — ZCode appends the path automatically. |
| I want to use the Claude / Anthropic mode | Switch the API format to Anthropic Messages (/v1/messages), and change the Base URL to https://api.haitoken.ai without /v1. This path should call POST /v1/messages. |
401 or authentication failed | Make sure the API key is valid and that ZCode is using OpenAI-compatible Bearer authentication. |
model not found | Make sure the model ID in ZCode exactly matches a model ID exposed by HaiToken. |
| Model was added but still cannot be used | After saving the provider, re-select the model once and, if needed, restart ZCode before testing again to avoid using a cached configuration. |
Next
