Third-party integration guide

HiLinkup can connect to OpenAI-, Anthropic/Claude-, and Gemini-style clients. Confirm the wire protocol and whether the field expects a base URL or a full endpoint before entering an address.

Connection values

Setting	Value
Provider type	OpenAI, Anthropic/Claude, or Gemini as required by the application
API key	A dedicated key from the HiLinkup console
Address	A base URL or full endpoint; these are not interchangeable
Model	An exact ID returned by `GET /v1/models`

Inspect the field label. A base URL is normally extended by the client; an endpoint normally requires the full resource path. Check the final URL for a missing or duplicated /v1.

Choose the correct protocol

Not every “compatible” client uses the same protocol. Check whether the application asks for OpenAI, Anthropic/Claude, or Gemini configuration.

Protocol	Use case	Base URL / endpoint	Auth	Reference
OpenAI Responses	Codex, Responses-capable agents, and clients that explicitly require this protocol	Base: `https://api.hilinkup.com/v1`; full: `.../v1/responses`	`Authorization: Bearer $HILINKUP_API_KEY`	Responses
OpenAI Chat Completions	WorkBuddy, traditional chat clients, `ChatOpenAI`, and chat-only plugins	Base: `https://api.hilinkup.com/v1`; full: `.../v1/chat/completions`	`Authorization: Bearer $HILINKUP_API_KEY`	Chat Completions
Claude Messages	Claude Code, Anthropic SDKs, Claude-style clients	Base: `https://api.hilinkup.com`; full: `.../v1/messages`	`x-api-key` plus `anthropic-version: 2023-06-01`	Claude Messages
Gemini generateContent	Gemini SDKs and Google AI-style clients	Base: `https://api.hilinkup.com`; full: `.../v1beta/models/{model}:generateContent`	`x-goog-api-key: $HILINKUP_API_KEY`	Gemini generateContent

Protocol names do not define model origin. Each protocol can call Chinese models on HiLinkup when that model and channel support the protocol. A model appearing in /v1/models does not by itself prove support for every protocol.

In /v1/models, owned_by identifies the provider and id uses provider/model format. Use the complete ID, such as aliyun/kimi-k3, to choose a provider explicitly. Sending only kimi-k3 uses the platform’s current default provider setting.

Common application mapping

Application or configuration	Protocol	Address
Codex CLI and Codex desktop	OpenAI Responses	Base URL `https://api.hilinkup.com/v1`
WorkBuddy custom Endpoint	OpenAI Chat Completions	Full endpoint `https://api.hilinkup.com/v1/chat/completions`
Claude Code and CC Switch Claude mode	Claude Messages	Final request must be `https://api.hilinkup.com/v1/messages`
Common chat clients and knowledge-base UIs	OpenAI Chat Completions	Usually base URL `https://api.hilinkup.com/v1`
LangChain `ChatOpenAI`, LlamaIndex `OpenAILike`, current Dify OpenAI-compatible setup	OpenAI Chat Completions	Usually base URL `https://api.hilinkup.com/v1`
Native Gemini clients	Gemini generateContent	Final request includes `/v1beta/models/{model}:generateContent`

Cursor, Cline, Roo Code, Continue, OpenCode, and OpenClaw change quickly. If the current version exposes separate Responses, Chat Completions, or Anthropic choices, follow that explicit protocol name instead of guessing from “OpenAI-compatible.”

Integration workflow

Create a dedicated API key

Use a separate key for each application and environment.

List models

curl https://api.hilinkup.com/v1/models \
  -H "Authorization: Bearer $HILINKUP_API_KEY"

Copy the complete provider/model value from data[].id, such as aliyun/kimi-k3.

Choose the protocol

Prefer Responses for new code. Use Chat Completions for traditional chat clients, Claude Messages for Claude Code, and generateContent for native Gemini clients.

Enter the correct address and authentication

Confirm whether the field expects a base URL or full endpoint, then verify the final path and headers.

Test plain chat

Send a short text message before enabling images, tools, knowledge bases, or long context.

Validate advanced capabilities separately

Test streaming, vision, tools, and structured output only with models that support them.

Supported integration guides

Developer tools

Claude Code, CC Switch, Codex CLI, the Codex desktop app, WorkBuddy, Cursor, Cline, Roo Code, Continue, OpenCode, and OpenClaw.

Chat clients

Cherry Studio, Chatbox, LobeChat, Open WebUI, NextChat, and LibreChat.

Knowledge bases

AnythingLLM and RAG-oriented workflows.

Frameworks

Dify, LangChain, Vercel AI SDK, and LlamaIndex.

Acceptance checklist

Connection test returns HTTP 200
The model is present in /v1/models
Plain text chat returns a non-empty response
Streaming completes normally
Usage appears in the console
Credentials are not exposed in client bundles, logs, or screenshots

cURL templates

Responses
OpenAI Chat
Claude
Gemini

curl https://api.hilinkup.com/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $HILINKUP_API_KEY" \
  -d '{
    "model": "aliyun/kimi-k3",
    "input": "Tell me a three sentence bedtime story about a unicorn."
  }'

curl https://api.hilinkup.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $HILINKUP_API_KEY" \
  -d '{
    "model": "aliyun/kimi-k3",
    "messages": [
      {
        "role": "developer",
        "content": "You are a helpful assistant."
      },
      {
        "role": "user",
        "content": "Hello!"
      }
    ]
  }'

curl https://api.hilinkup.com/v1/messages \
  -H "content-type: application/json" \
  -H "x-api-key: $HILINKUP_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "aliyun/kimi-k3",
    "max_tokens": 1024,
    "messages": [
      {
        "role": "user",
        "content": "Hello, Claude"
      }
    ]
  }'

curl "https://api.hilinkup.com/v1beta/models/aliyun/kimi-k3:generateContent" \
  -H "x-goog-api-key: $HILINKUP_API_KEY" \
  -H "Content-Type: application/json" \
  -X POST \
  -d '{
    "contents": [
      {
        "parts": [
          {
            "text": "Find the race condition in this multi-threaded C++ snippet: [code here]"
          }
        ]
      }
    ]
  }'

Common failures

cURL works but the application fails

Inspect the final protocol, URL, and authentication header. Responses uses /v1/responses, Chat Completions uses /v1/chat/completions, Claude uses /v1/messages, and Gemini uses /v1beta/models/{model}:generateContent.

The model list is empty

Add an exact model ID manually. Some clients cannot discover models from custom providers.

Chat works but vision or tools fail

Verify model capability and the application’s model capability flags. Authentication is already working.

Overview

Developer tools

Chat clients

Application frameworks

Third-party integration guide

Connection values

Choose the correct protocol

Common application mapping

Integration workflow

Supported integration guides

Developer tools

Chat clients

Knowledge bases

Frameworks

Acceptance checklist

cURL templates

Common failures

​Connection values

​Choose the correct protocol

​Common application mapping

​Integration workflow

​Supported integration guides

Developer tools

Chat clients

Knowledge bases

Frameworks

​Acceptance checklist

​cURL templates

​Common failures

Connection values

Choose the correct protocol

Common application mapping

Integration workflow

Supported integration guides

Acceptance checklist

cURL templates

Common failures