ai大模型调用协议

传输层

1. http/https
2. sse
3. websocket
4. grpc

目前主要的大模型api调用使用http+sse实现请求建立与大模型回答的增量推送。

api规范层

1. openAI 协议

协议	端点	会话模型	特点
Chat Completions	POST /chat/completions	无状态messages[]数组	每次需要携带完整上下文，兼容性最广
Response API	POST /responses	有状态	新协议，支持长时agent，内置工具调用，推理过程独立输出

Chat Completions示例：

curl -L -X POST 'https://api.deepseek.com/chat/completions' \
-H 'Content-Type: application/json' \
-H 'Accept: application/json' \
-H 'Authorization: Bearer <TOKEN>' \
--data-raw '{
  "messages": [
    {
      "content": "You are a helpful assistant",
      "role": "system"
    },
    {
      "content": "Hi",
      "role": "user"
    }
  ],
  "model": "deepseek-flash",
  "thinking": {
    "type": "enabled"
  },
  "reasoning_effort": "none",
  "max_tokens": 4096,
  "response_format": {
    "type": "text"
  },
  "stop": null,
  "stream": false,
  "stream_options": null,
  "temperature": 1,
  "top_p": 1,
  "tools": null,
  "tool_choice": "none",
  "logprobs": false,
  "top_logprobs": null
}'

Response API 示例：

curl -L -X POST 'https://api.deepseek.com/responses' \
-H 'Content-Type: application/json' \
-H 'Accept: application/json' \
-H 'Authorization: Bearer <TOKEN>' \
--data-raw '{
  "model": "deepseek-flash",
  "input": "string",
  "instructions": "string",
  "reasoning": {
    "effort": "none"
  },
  "max_output_tokens": 4096,
  "stream": false,
  "temperature": 1,
  "top_p": 1,
  "text": {
    "format": {
      "type": "text",
      "name": "string",
      "schema": {}
    }
  },
  "tools": null,
  "tool_choice": "none",
  "top_logprobs": null,
  "user": "string"
}'

2. Anthropic Messages 协议

• 端点：POST /v1/messages
• 结构：system顶层 + messages[]（仅user/assistant交替），content为block数组
• 工具调用：通过tool_use block和tool_result block实现
• 思考能力：thinking: {budget_tokens}，思考过程作为独立content block输出

3. Google Gemini 原生协议

• 端点：/gemini/v1beta/models/{model}:generateContent（非流式）和:streamGenerateContent（流式）
• 认证：使用x-goog-api-key Header
• 特点：模型名称置于URL路径中，而非请求体内

4. 其他协议

agent协作层

• mcp
• A2A
• AG-UI