Third-Party Tool Integration

Applicable role: Developer Last updated: 2026-08-06

Any client that supports a custom Base URL can typically call models through the platform. Field names vary slightly across tools, but the configuration is essentially the same: API endpoint, API Key, model call name, and protocol type.

Step-by-step instructions for specific tools are available in the Tool Integration Guide. This page explains how to choose the correct endpoint for each type of tool and how to verify that your requests are properly routed after configuration.

Prepare Your Connection Details

Go to the API Key page in the console, then create or select a key. You can copy model call names from the Call Guide. The examples below use deepseek-v4-flash; if you need kimi-k3, copy the corresponding call name from the call guide.

Most tools only require the Base URL and will automatically append /chat/completions or /messages. Only fill in the full URL when the field is explicitly labeled Endpoint, API URL, or full endpoint address.

Choose Configuration by Tool Type

OpenAI Compatible Clients

Tools such as Cursor, Cline, Roo Code, Continue, Chatbox, Cherry Studio, Open WebUI, Dify, and Coze all provide an OpenAI Compatible or Custom OpenAI entry. The recommended configuration is:

Provider: OpenAI Compatible
Base URL: https://<your-api-endpoint>
API Key: YOUR_API_KEY
Model: deepseek-v4-flash

The client appends the request path based on the protocol. If the final request URL contains a duplicated path, verify that you entered the root domain address above rather than the full endpoint URL.

Messages Compatible Clients

Claude Code, Hermes Agent, and some agent frameworks use the Messages format. These tools typically accept the root domain address:

Base URL: https://<your-api-endpoint>
API Key: YOUR_API_KEY
Model: deepseek-v4-flash

The tool appends /v1/messages when making requests. If the client requires a full endpoint, enter https://<your-api-endpoint>/v1/messages.

Tools That Require a Full Endpoint URL

Immersive Translate, generic HTTP nodes, and some low-code platforms do not automatically append the endpoint path. Use:

https://<your-api-endpoint>/v1/chat/completions

The authentication header is:

Authorization: Bearer YOUR_API_KEY

Tool Directory

Verify the Connection

After saving your configuration, send a short test message:

Reply with only: connection successful

Then check the following:

1. The client receives a normal response with no 401, 403, or 404 errors.
2. A new entry appears under Usage Records in the console.
3. The key, model, and group in the record match your configuration.
4. For agent-type tools, run a simple tool call to confirm that file read/write or command execution capabilities work correctly.

If chat responses work normally but the agent cannot invoke tools, the issue is usually not a connection problem. Check whether the model supports tool calling and whether the client has enabled the corresponding capability.

Common Issues

401 Returned

Re-copy your API Key and make sure the input field does not contain spaces, line breaks, or a duplicate Bearer prefix. Most API Key fields only accept the sk-xxx value.

403 Returned

Check whether the key is enabled, whether it has expired, whether it is bound to an available group, and whether the current user has access to that group. Insufficient balance or quota may also return 403; refer to the code field in the response for details.

404 Returned

Inspect the final request URL sent by the client. Common causes include the client not appending the path according to the protocol, or pasting the full endpoint URL into a field that only accepts a Base URL.

Model Not Found Returned

Re-copy the model name from the Call Guide in the console. The display name and the call name may differ; always use the call name in your configuration.

Requests Still Go to the Original Service After Configuration

Fully quit and restart the client, then check system environment variables, user-level configuration, and project-level configuration. Project-level configuration typically overrides user-level configuration.

Key Management

Use separate keys for different team members and applications, and set quota, rate, and source restrictions for each. Do not commit configuration files containing keys to code repositories. When a key is no longer in use or you suspect it has been compromised, disable it in the console promptly and create a new one.

For more API details, see API Basics. For error responses and retry strategies, see Error Codes and Troubleshooting.