大模型供应商API端点兼容协议

各家大模型供应商为了兼容各种AI工具基本上都会支持几种广泛使用的API兼容协议一、常见的兼容协议1.OpenAI Chat Completions(v1/chat/completions) 协议请求示例curl --location https://open.example.com/v1/chat/completions \ --header Authorization: Bearer sk-aaabbb \ --header Content-Type: application/json \ --data { model: GLM-5.2, messages: [ {role: user, content: hello} ], temperature: 0.2, top_p: 0.9 }返回示例{ id: 5aa825e5-3yud-4il0-bfhj-c619c3c39367, object: chat.completion, model: glm-5.2, created: 1785480756, choices: [ { index: 0, message: { role: assistant, content: 你好我是Z.ai训练的GLM大语言模型。很高兴能与你交流。\n\n有什么我能帮助你的问题或者你想了解的话题吗无论是工作学习上的疑问还是日常生活中的好奇我都很乐意为你提供帮助。, reasoning_content: Let me consider how to respond to this greeting thoughtfully.\n\nThe user has started with a simple \hello\ - this is likely the beginning of a conversation. I should respond in a way thats both welcoming and open-ended.\n\nFirst, Ill acknowledge their greeting warmly. Then I should introduce myself briefly to establish context. Since this is an initial interaction, I should also express my willingness to help while inviting them to share whats on their mind.\n\nI need to make sure my response feels natural and conversational, not overly formal or robotic. It would be good to end with an open question that encourages them to continue the dialogue.\n\nThe key is to create a friendly atmosphere where they feel comfortable sharing what they need assistance with. This initial exchange sets the tone for our entire conversation, so its important to get it right.\n\nLet me craft a response thats welcoming, informative, and encourages further conversation. }, finish_reason: stop } ], usage: { prompt_tokens: 13, completion_tokens: 232, total_tokens: 245, prompt_tokens_details: { cached_tokens: 0 }, completion_tokens_details: { reasoning_tokens: 183 } } }协议由美国 OpenAI 公司原生开发2023 年 3 月伴随 gpt-3.5-turbo 上线替代老式单轮 /v1/completions2.Anthropic Messages(v1/messages)协议请求示例curl --location https://open.example.com/v1/messages \ --header x-api-key: sk-aaabbb \ --header anthropic-version: 2023-06-01 \ --header Content-Type: application/json \ --data { model: GLM-5.2, max_tokens: 2048, temperature: 0.2, top_p: 0.9, messages: [ {role: user, content: hello} ] }返回示例{ id: 390f957a-09ee-40cf-845e-6af58a34aa0e, type: message, role: assistant, model: glm-5.2, content: [ { type: text, text: Hello! Im GLM, trained by Z.ai. How can I assist you today? Whether you have questions or just want to chat, Im here to help. } ], stop_reason: end_turn, stop_sequence: null, usage: { input_tokens: 13, output_tokens: 272 } }协议由由美国 Anthropic 公司原生独立设计开发正式发布时间2023 年 6 月API 版本头 anthropic-version: 2023-06-01✅ 最早应用模型Claude 2 / Claude 2.1推出目的替代老旧字符串 prompt 接口支持结构化多轮对话、区分 user/assistant 消息轮次架构关键设计system 独立顶层字段不塞进 messages 数组强制消息交替 user ↔ assistantmax_tokens 为必填项。3.OpenAI Responses API/v1/responses协议(CodeX使用)请求示例curl --location https://open.example.com/v1/responses \ --header Authorization: Bearer sk-aaabbb \ --header Content-Type: application/json \ --data { model: GLM-5.2, input: [ { role: user, content: hello } ], temperature: 0.2, top_p: 0.9 }返回示例{ id: 4f4f58e3-c6bf-48bb-a1cc-fc25375dd1f3, object: response, created_at: 1786443315, model: glm-5.2, status: completed, output: [ { type: message, id: msg_4dee0eac9d29492e9d1d04cbae1cc8e1, role: assistant, status: completed, content: [ { type: output_text, text: 你好我是Z.ai训练的GLM大语言模型。很高兴能与你交流。\n\n有什么我能帮助你的问题或者你想了解的话题吗无论是工作学习上的疑问还是日常生活中的好奇我都很乐意为你提供帮助。, annotations: [] } ] } ], usage: { input_tokens: 13, output_tokens: 232, total_tokens: 245 } }协议由美国 OpenAI 独立原生开发定位OpenAI 面向智能体 (Agent) 场景的新一代官方 REST 协议正式发布2025-03-11OpenAI 官方对外推出 /v1/responses设计目标替代老旧 Beta 版 Assistants API并作为 Chat Completions 的下一代演进接口。最早应用新版 Codex 云编码智能体codex-14.Google Gemini(/v1beta/models)协议请求示例curl --location http://localhost:8787/v1beta/models/GLM-5.2:generateContent \ --header Content-Type: application/json \ --header x-goog-api-key: sk-aaabbb \ --data { contents: [ { role: user, parts: [{text: hello}] } ] }响应示例{ candidates: [ { content: { role: model, parts: [ { text: Hello! Im GLM, trained by Z.ai. How can I assist you today? Whether you have questions or just want to chat, Im here to help. } ] }, finishReason: STOP, index: 0 } ], usageMetadata: { promptTokenCount: 13, candidatesTokenCount: 192, totalTokenCount: 205 }, modelVersion: glm-5.2 }二、各厂商兼容情况各厂商和中转站采用new-api、CCX、TokenHub等开源组件搭建基本上都支持前两种少量情况也会支持第三种比如稀宇科技minimax的官方文档里提到了支持三种智谱AI的官方文档提到了支持前两种三、各编码工具对接情况1.claude命令行工具修改配置文件 C:\Users\[用户名]\.claude\settings.json内容如下{ env: { ANTHROPIC_API_KEY: sk-aaabbb, ANTHROPIC_AUTH_TOKEN: sk-aaabbb, ANTHROPIC_BASE_URL: https://open.example.com, ANTHROPIC_DEFAULT_OPUS_MODEL: GLM-5.2, ANTHROPIC_DEFAULT_SONNET_MODEL: GLM-5.2, ANTHROPIC_DEFAULT_HAIKU_MODEL: GLM-5.2, ANTHROPIC_MODEL: GLM-5.2, ANTHROPIC_REASONING_MODEL: GLM-5.2 } }2.vscode安装claude code for vscode插件配置同上3.jetbrain idea3.1 proxyAI插件安装 proxyAI插件 只能对话不能直接修改代码配置导入CustomOpenAiSettings.json内容[{name:customendpoint,template:OPENAI,apiKey:,chatCompletionSettings:{url:https://open.example.com/v1/chat/completions,headers:{Authorization:Bearer $CUSTOM_SERVICE_API_KEY,X-LLM-Application-Tag:proxyai,Content-Type:application/json},body:{stream:true,model:GLM-5.2,messages:$OPENAI_MESSAGES,temperature:0.2,max_tokens:8192}},codeCompletionSettings:{codeCompletionsEnabled:true,parseResponseAsChatCompletions:true,infillTemplate:OPENAI,url:https://open.example.com/v1/chat/completions,headers:{Authorization:Bearer $CUSTOM_SERVICE_API_KEY,X-LLM-Application-Tag:proxyai,Content-Type:application/json},body:{suffix:$SUFFIX,stream:true,model:GLM-5.2,temperature:0.2,prompt:$PREFIX,max_tokens:1024}}}]3.2 Tencent Cloud CodeBuddy插件自带每月免费模型额度 高峰期排队可以直接修改代码现阶段支持OpenAI Chat Completions(v1/chat/completions) 协议的URLAuto -- 配置自定义模型供应商选other-custom BASE URLhttps://open.example.com/v1/chat/completions API KEY你的密钥 模型ID: GLM-5.2 高级设置: 勾选 工具调用 和 推理4.trae IDE基于vscode的编码ide 自带免费模型可以直接修改代码2026.7.31改成每月免费500积分了点击 右上角齿轮 进入设置页面点击 模型--添加模型点击 自定义配置 设置如下API格式Anthropic Message格式 自定义请求地址https://open.example.com 模型ID: GLM-5.2 API密钥: 你的密钥TRAE任务栏 模型选择关闭auto-mode 改成刚才定义的GLM-5.2目前trae支持 OpenAI Chat Completions(v1/chat/completions) 协议和Anthropic Messages(v1/messages)协议5.Zcode交互式编程助手 类似codex 可以编码也可以通过技能实现写ppt word可以在不登录的情况下配置自定义模型点击右下角齿轮-模型设置-添加供应商设置如下名称 customendpoint baseurl https://open.example.com api key 你的api key api格式 Anthropic Message格式 添加模型 模型id填入 GLM-5.2返回工作区即可使用刚才配置的模型id了zcode支持 OpenAI Chat Completions(v1/chat/completions) 协议和Anthropic Messages(v1/messages)协议6.Hermes Agent由 Nous Research 开发Github 高星自进化 AI Agent 工具核心能力闭环自我学习自动把重复工作提炼为可复用 Skill 技能后续任务提速 40%内置长效记忆系统自动清理无用对话、留存长期规则自定义endpoint支持OpenAI Chat Completions(v1/chat/completions) 协议、Anthropic Messages(v1/messages) 协议 和 OpenAI Responses API/v1/responses协议(Codex使用)交互式命令切换模型hermes model填写 如下Select provider: 选择 (○) Custom endpoint (enter URL manually) API base URL [e.g. https://api.example.com/v1]: https://open.example.com/v1 API key [optional]: 你的api key Select API compatibility mode: 选择2 Chat Completions Available models: 选择2 GLM-5.2 Context length in tokens: 设置上下文长度 直接回车自动检测即可 Display name: 回车使用命令行hermes切换模型hermes model7.NetcattyNetcatty 是一款跨平台的ssh工具支持接入OpenAI Chat Completions协议的模型设置-AI-提供商-添加提供商显示名称customendpoint API key: 你的api key Base URL : https://open.example.com/v1 默认模型 GLM-5.2保存后点击启用在连接主机后 点击右上角AI按钮 可以通过对话使用最新版本的Netcatty已经支持 OpenAI Chat Completions(v1/chat/completions) 协议和Anthropic Messages(v1/messages)协议 额外支持google gemini协议8.OpsKatOpsKat 一款AI 一站式服务器运维桌面客户端可以统一管理 Linux 服务器、数据库、K8s 集群、Redis/MQ 中间件内置 SSH、SFTP、终端、审计日志还能接入大模型自然语言执行运维命令。设置-AI-提供商-添加提供商porvider类型 OpenAI兼容 名称customendpoint 模型 GLM-5.2 API 地址 : https://open.example.com/v1 API key: 你的密钥 其他默认