Tool Integration Guide
Applicable roles: Developers, Admin Users, Member Users Last updated: 2026-08-06
When integrating a third-party tool with the platform, the key is to determine whether it requires a Base URL or a full request URL. Field names may vary across tool versions, but the actual connection information consists of only the following items.
Prepare These Four Items First
| Configuration | Value |
|---|---|
| API Key | An API key starting with sk-, created in the console |
| Model call name | Copy from the console "Call Guide", e.g. deepseek-v4-flash |
| OpenAI Compatible Base URL | https://<your-api-endpoint> |
| Messages Compatible Base URL | https://<your-api-endpoint> |
Use the API endpoint shown on the console "API Key" page.
Create your API key on the "API Key" page in the left sidebar of the console. After creating it, you must also bind a group to it under "API Key -> Edit -> Route Group": a key without a bound group will return 403 when called with a model name, with the error message API Key is not assigned to any group and cannot be used.
When a field is labeled Base URL, API Base, or Provider URL, you typically enter the Base URL. When a field explicitly says Endpoint, Request URL, or full request address, enter the following for OpenAI Compatible requests:
https://<your-api-endpoint>/v1/chat/completionsFor Messages Compatible requests, enter:
https://<your-api-endpoint>/v1/messagesDo not enter the full request URL in a Base URL field, as the client may append the path again, resulting in a duplicated endpoint. Use the root address from the table above as your Base URL.
Choose a Tutorial by Use Case
Code and Terminal
| Tool | Integration Method | Tutorial |
|---|---|---|
| Claude Code | Messages Compatible | Claude Code |
| Codex CLI | Custom Provider | Codex CLI |
| Aider | OpenAI Compatible | Aider |
| OpenCode | OpenAI Compatible Provider | OpenCode |
| Qwen Code / CodeWhale | Custom Provider (depends on current version) | Qwen Code and CodeWhale |
| LangChain / LiteLLM | SDK parameters | LangChain and LiteLLM |
IDEs and Coding Plugins
| Tool | Integration Method | Tutorial |
|---|---|---|
| Cursor | Custom OpenAI Base URL (affected by version capabilities) | Cursor |
| Cline | OpenAI Compatible Provider | Cline |
| Roo Code / Kilo Code | OpenAI Compatible Provider | Roo Code / Kilo Code |
| Continue | Local YAML configuration | Continue |
| Trae / Zed / Windsurf | Native entry or compatible plugin | Trae, Zed, and Windsurf |
Chat, Agents, and Workflows
| Type | Tools and Tutorials |
|---|---|
| Desktop and Web Chat | Chatbox, Cherry Studio, Open WebUI, LobeChat and NextChat |
| Messaging Channels and Agents | ChatGPT-on-WeChat, OpenClaw, Hermes Agent |
| Applications and Automation | Dify and Coze, n8n and Flowise |
| Web Translation | Immersive Translate |
How to Verify a Successful Integration
Send a request containing only a short text message first, then check the console "Usage Records". If the API key, model, and group in the records match your tool configuration, the request has been routed through the platform.
For coding agents, you should also run a tool-call test, such as reading a file that contains no sensitive information. If chat works normally but file operations fail, the connection is typically successful, but the model capabilities, tool permissions, or client settings may need adjustment.
For 401 errors, check the key first. For 403 errors, first verify whether the API key has a bound route group (an unbound group is the most common cause), then check the key status, group permissions, and billing source. For 404 errors, check the final request path and model call name. For detailed troubleshooting, see Error Codes and Troubleshooting.
Images in this directory are from the actual interfaces of the corresponding software. To align with this documentation, the service name, API endpoint, and model call name in the screenshots have been replaced with the platform's configuration, and sensitive information such as API keys has been redacted. Menu names and positions may change across software versions.