Third-Party Tool Integration

Applicable role: Developer Updated: 2026-07-14

The platform is compatible with the OpenAI / Anthropic protocols, so it can integrate with the vast majority of AI tools and frameworks that support a "custom Base URL." This article covers the configuration methods for mainstream tools across three categories: command-line tools → AI coding IDEs/plugins → application development frameworks.

The API address is YOUR_API_BASE_URL, the route prefix is /v1, and authentication uses Authorization: Bearer sk-xxx.

Quick reference for what to fill in: for the vast majority of tools you only need three things—① Base URL: YOUR_API_BASE_URL; ② API Key: the platform's sk--prefixed Key (created and distributed by your organization in the console); ③ Model name: the model name (e.g. claude-sonnet-4-6, available in the console's "Usage Guide" or "Model Marketplace"). If your organization has enabled group-tag routing (an advanced feature, disabled by default), the Usage Guide page also provides a group-tagged call name (e.g. 5MHXZWKA/gpt-4o) to precisely target a group—see Usage Guide and Routing for details. The actual path for the Anthropic protocol endpoint is /v1/messages: for tools using the official SDK (such as Claude Code), simply set the Base URL to the domain root YOUR_API_BASE_URL, and the SDK will automatically append /v1/messages; if a tool requires you to fill in the complete Base URL directly, then use YOUR_API_BASE_URL. For tools not listed in this article, simply apply the General Integration Principles.

Command-Line Tools (CLI)

1. Claude Code

Claude Code is Anthropic's official CLI tool, and it injects the platform configuration via environment variables.

Set the following in your shell configuration file (~/.zshrc or ~/.bashrc):

export ANTHROPIC_BASE_URL="YOUR_API_BASE_URL"
export ANTHROPIC_API_KEY="YOUR_API_KEY"

After saving, run source ~/.zshrc to apply the changes, then use the claude command as usual.

Model tiers: In Claude Code, use /model to switch between the Opus / Sonnet / Haiku tiers. The platform routes to the corresponding model according to the mapping. For the specific available models, refer to the console's "Usage Guide" or "Model Marketplace."

Tip: For pricing, refer to what is shown in the "Model Marketplace."

2. Codex CLI

OpenAI Codex CLI uses the OpenAI-compatible protocol and is configured via environment variables:

export OPENAI_BASE_URL="YOUR_API_BASE_URL"
export OPENAI_API_KEY="YOUR_API_KEY"

Alternatively, specify the base_url and model name of a custom provider in ~/.codex/config.toml, then launch with that provider. Model names can be found in the "Model Marketplace"; the model name alone is sufficient. Once group-tag routing is enabled, you can also use the group-tagged call name.

3. Aider

Aider natively supports OpenAI-compatible endpoints. Using environment variables is recommended:

export OPENAI_API_BASE="YOUR_API_BASE_URL"
export OPENAI_API_KEY="YOUR_API_KEY"
 
# Add the openai/ prefix to the model name to use the compatible endpoint
aider --model openai/claude-sonnet-4-6

You can also use the command-line flags --openai-api-base / --openai-api-key, or write them into .aider.conf.yml.

⚠️ Key point: When pointing to a compatible endpoint, the model name must include the openai/ prefix, otherwise Aider will resolve it against a built-in provider. If using a group-tagged call name, the format looks like openai/5MHXZWKA/gpt-4o (Aider uses the openai/ part to select the provider and strips it off, passing 5MHXZWKA/gpt-4o through to the platform). It's recommended to test this three-segment format with a small amount of traffic first; if the group tag doesn't take effect, fall back to the model name openai/claude-sonnet-4-6 (which uses the routing group bound to this Key).

AI Coding IDEs / Plugins

The following VSCode-family plugins (Cline / Roo Code / Kilo Code / Continue, etc.) have highly consistent configuration: in the "OpenAI Compatible / custom provider" section, simply fill in the Base URL + API Key + model call name. Model names can be found in the "Model Marketplace"; the model name alone is sufficient. Once group-tag routing is enabled, you can also use the group-tagged call name.

4. Cursor

Cursor supports custom OpenAI API configuration.

1. Open Cursor Settings → Models
2. Find the OpenAI API Key configuration item and enter the platform's API Key
3. Expand Override OpenAI Base URL and enter: YOUR_API_BASE_URL
4. Add the call name provided in the Usage Guide to the model list (either the model name or the group-tagged form), then save and verify the connection

Cursor shares the same origin as VSCode, so its configuration system is consistent.

5. Cline (VSCode plugin)

1. Click the settings icon in the top-right corner of the Cline panel
2. Set API Provider to OpenAI Compatible
3. Set Base URL to YOUR_API_BASE_URL
4. Set API Key to the platform's Key
5. Set Model ID to the call name provided in the Usage Guide (e.g. claude-sonnet-4-6, or the group-tagged 5MHXZWKA/gpt-4o)

💡 If you primarily use Claude models and want full native tool calling (reading/writing files, executing commands), you can also set the Provider to Anthropic with the Base URL as YOUR_API_BASE_URL (without /v1).

6. Roo Code / Kilo Code (VSCode plugins)

Both are Cline derivatives and share the same configuration process:

1. Open the plugin settings panel (gear icon)
2. Set API Provider to OpenAI Compatible
3. Fill in Base URL YOUR_API_BASE_URL, API Key, and Model (the call name is in the Usage Guide; either the model name or the group-tagged form works)
4. Save

⚠️ Roo Code only supports native tool calling, so please choose a model that supports tool calling. Kilo Code's Base URL supports filling in the full path (including /chat/completions), making it more flexible.

7. Continue (VSCode / JetBrains plugin)

Edit ~/.continue/config.yaml to use the platform as a provider: openai endpoint:

models:
  - name: PlatformSonnet
    provider: openai
    model: claude-sonnet-4-6
    apiBase: YOUR_API_BASE_URL
    apiKey: YOUR_API_KEY
    roles:
      - chat
      - edit

For model, fill in the call name provided in the Usage Guide (either the model name or the group-tagged form). When multiple models share the same Base URL / Key, you can use YAML anchors (& / <<: *) to reduce repetition.

8. Zed Editor

Zed natively supports custom OpenAI-compatible providers. Edit settings.json:

{
  "language_models": {
    "openai_compatible": {
      "my-platform": {
        "api_url": "YOUR_API_BASE_URL",
        "available_models": [
          {
            "name": "claude-sonnet-4-6",
            "display_name": "Platform Sonnet",
            "max_tokens": 128000
          }
        ]
      }
    }
  }
}

For name, fill in the call name provided in the Usage Guide (either the model name or the group-tagged form). The API Key is entered in the provider settings UI (stored in the system keychain), or set the environment variable YOUR_PLATFORM_API_KEY (generated from the uppercase provider ID plus _API_KEY).

⚠️ api_url should not have a trailing slash, and it must correctly include /v1.

9. Windsurf

Windsurf's native BYOK only supports Anthropic models. To integrate the platform's OpenAI-compatible endpoint, we recommend installing any of the Cline / Roo Code / Continue plugins within Windsurf, then configuring the Base URL + Key according to the corresponding section above.

If your version of Windsurf exposes a custom Base URL input field, you can enter YOUR_API_BASE_URL directly; if it only accepts the official Key and has no Base URL input field, that version only supports official BYOK, and you'll need to use the plugin approach.

10. Other VSCode-Family Tools

Editors such as Void and Tabby that support "OpenAI-compatible provider / custom Base URL" follow the same configuration logic as the plugins above: in their provider settings, simply fill in the Base URL YOUR_API_BASE_URL, the API Key, and the model call name.

Application Development Frameworks

11. LangChain

Use ChatOpenAI to point to the platform:

from langchain_openai import ChatOpenAI
 
llm = ChatOpenAI(
    model="claude-sonnet-4-6",
    api_key="YOUR_API_KEY",
    base_url="YOUR_API_BASE_URL"
)
 
print(llm.invoke("你好").content)

Install: pip install langchain-openai

12. LiteLLM

LiteLLM uses the platform as an OpenAI-compatible provider:

import litellm
 
resp = litellm.completion(
    model="openai/claude-sonnet-4-6",          # openai/ prefix uses the compatible endpoint
    api_key="YOUR_API_KEY",
    base_url="YOUR_API_BASE_URL",
    messages=[{"role": "user", "content": "你好"}]
)
print(resp.choices[0].message.content)

Install: pip install litellm

13. Dify

In Dify, integrate the platform as an "OpenAI-API-compatible" model provider:

1. Go to Dify → Settings → Model Providers
2. Select OpenAI-API-compatible
3. Fill in:
   • API Key: the platform's Key
   • API Base URL: YOUR_API_BASE_URL
   • Model Name: the call name provided in the Usage Guide (e.g. claude-sonnet-4-6, or the group-tagged form)
4. After saving, you can select this model in Dify applications

14. n8n / Other Low-Code Platforms

Low-code/workflow platforms such as n8n and Flowise mostly include built-in "OpenAI" or "OpenAI compatible" nodes. Simply change the Base URL in the node credentials to YOUR_API_BASE_URL, fill in the platform's Key as the API Key, and set the model name to the call name provided in the Usage Guide.

General Integration Principles

Any tool that supports the following configuration can integrate with the platform:

Common Troubleshooting:

• 401 / Invalid API Key → Check whether the Key was entered incorrectly or contains extra whitespace
• 404 / Model Not Found → Check whether the model call name is a valid call name from the Usage Guide
• "Group unreachable" with a group tag → The group is not within your available range or is currently unavailable; see Usage Guide and Routing
• Connection failure → Check whether the Base URL is correct, whether /v1 is missing or duplicated, and whether there is an extra trailing slash

You only need to change one API address and Key—no changes to any business code are required. This is precisely the value of the platform's compatibility.

Next Steps

• Understand how to write the model field and the routing rules → Usage Guide and Routing
• Encountering errors → Error Codes and Troubleshooting
• Model occasionally outputs a foreign language → Claude Language Drift Notes