比邻
4家人工智能提供商。1台本地服务器。没有API密钥。
直接在您的编码工具中使用ChatGPT、Claude、Gemini和Perplexity——通过您现有的帐户。
入门指南 · 命令行界面 · REST API · 双向通信 · SDK · MCP工具
______________________________________________________________________
演示
应用程序演示·CLI·Webhook实时聊天和战斗·应用程序概述
https://github.com/user-attachments/assets/5e75eb68-b1b5-43dc-979d-3bf6faa48fa0
https://github.com/user-attachments/assets/a8564fc9-b3b3-4a53-bc35-cfce72fe34da
https://github.com/user-attachments/assets/bb7fa455-d379-4e69-b530-f7c09d2faccf
https://github.com/user-attachments/assets/d4121fdb-f97e-4d35-846c-5ec7c5249a85
______________________________________________________________________
💖 赞助商墙
*Proxima不断发展，这要归功于这些了不起的人*
@TheNetworker
⭐ Star Sponsor · 🥇 First Sponsor
*伟大的事物是共同构建的--* 成为我们旅程的一部分→
______________________________________________________________________
概述
Proxima是一个本地AI网关，将多个AI提供商连接到您的开发环境。它通过您的活动登录会话在浏览器级别与每个提供商进行通信，就像您在浏览器中与他们聊天一样。
🌐 One Endpoint Everything through /v1/chat/completions — no separate URLs
🤖 4 AI Providers ChatGPT, Claude, Gemini, Perplexity — any model, any task
⚡ Provider Engines Native browser-level communication — 3–10x faster, more reliable
🖥️ CLI Tool proxima ask, proxima fix, proxima debate — right from your terminal
🔌 WebSocket Real-time streaming at ws://localhost:3210/ws
🧰 45+ MCP Tools Search, code, translate, analyze, debate, audit — all via MCP
📡 REST API OpenAI-compatible API on localhost:3210
📦 SDKs Python & JavaScript — one function each
🧠 Smart Router Auto-picks the best available AI for your query
🔑 No API Keys Uses your existing browser sessions — see how it works
🔒 Local & Private Runs on 127.0.0.1, data goes only to providers you're logged into
______________________________________________________________________
v4.1.0的新增功能
🔥 Provider Engine System Proxima now uses native browser-level communication with AI providers — no DOM scraping. Responses are 3–10x faster and far more stable, with SSE streaming support and automatic fallback mechanisms.
⚡ CLI Tool Run proxima ask, proxima fix, proxima debate from any terminal. Pipe errors straight from your build output. Supports file context, git diff piping, and JSON output for scripts.
🔌 WebSocket Server Real-time streaming AI at ws://localhost:3210/ws. Bidirectional communication with status updates, request tracking, and keepalive. Useful for apps, scripts, anything that needs live output.
🛠️ 15 New MCP Tools chain_query, solve, debate, security_audit, verify, fix_error, build_architecture, write_tests, explain_error, convert_code, ask_selected, conversation_export, ask_perplexity, github_search, get_ui_reference
📄 Interactive API Docs Live documentation at /docs, /cli, /ws — with a working chat widget to test queries directly in your browser.
🎯 Multi-Model Queries model: "all" queries every provider at once. model: ["claude", "chatgpt"] targets specific ones. Compare responses side-by-side from multiple AI providers in a single request.
📤 Conversation Export Export full conversation history from any provider using conversation_export. Continue working on AI agent projects, revisit ideas discussed with providers, and build on previous plans without losing context.
🛡️ New REST API Functions New security_audit and debate functions added to the REST API endpoint. File upload support via file field in request body.
Bug修复和改进：
- 🔧 交错的多提供者查询——防止并行请求期间UI冻结
- 🔧 明智的供应商选择——将编码任务交给克劳德，将研究交给困惑
- 🔧 具有TTL（5分钟）和自动驱逐（最多100个条目）的响应缓存
- 🔧 速率限制处理——检测429个响应，在过期会话上自动恢复
- 🔧 发动机自动喷射页面导航，带双重防护
- 🔧 Claude对话自动恢复（处理404/410过期会话）
- 🔧 ChatGPT SHA3-512工作量证明挑战求解器
- 🔧 带有CORS头的REST API上的10MB主体大小限制
- 🔧 IPC重新连接时防止插座泄漏
______________________________________________________________________
入门指南
需求
- （适用于MCP服务器和CLI）
- Windows 10/11 --预构建安装程序可用
- macOS/Linux --通过源代码支持
安装
下载安装程序（Windows）
下载最新版本并运行安装程序。
从源代码运行（Windows/macOS/Linux）
git clone https://github.com/Zen4-bit/Proxima.git
cd Proxima
npm install
npm start电子将打开Proxima窗口。登录到您的人工智能提供商，在设置中启用REST API，您就准备好了。
CLI安装：
- 窗户： 设置→ ⚡ 将CLI安装到PATH，或
npm link - macOS/Linux：
npm link（可能需要sudo npm link)
连接到编辑器
- 打开Proxima并登录您的AI提供商（一次性设置）
- 首选 设置→ MCP配置 → 复制配置
- 粘贴到编辑器的MCP配置文件中：
{
"mcpServers": {
"proxima": {
"command": "node",
"args": ["C:/path/to/Proxima/src/mcp-server-v3.js"]
}
}
}- 重新启动编辑器。工具将会出现。
提示： 使用设置中的复制按钮——不要手动键入路径。
适用于： 光标·VS代码（MCP扩展）·克劳德桌面·Windsurf·Gemini CLI·任何兼容MCP的客户端
______________________________________________________________________
支持的提供商
ChatGPT
OpenAI's GPT
Claude
Anthropic's Claude
Gemini
Google's Gemini
Perplexity
Web search & research
每个供应商都通过一个专门的 引擎脚本 它在浏览器级别处理通信。使用您现有的登录名通过SSE流式传输响应。如果引擎无法连接，Proxima会自动退回到基于DOM的交互。
______________________________________________________________________
运作原理
在v4.1.0中，Proxima使用 提供商引擎系统 而不是DOM抓取。
当您发送查询时，Proxima在提供商的浏览器选项卡中使用轻量级引擎脚本。该脚本在浏览器级别处理通信，并通过SSE流式传输响应。如果引擎因任何原因发生故障，Proxima会自动回退到基于DOM的交互，因此它可以继续以任何方式工作。
Your editor → MCP tool call → Proxima local server
↓
Engine injected into session
↓
Browser-level communication (SSE stream)
↓
Response returnedEngineProviderHow it works chatgpt-engine.jsChatGPTHandles proof-of-work challenges, streams via SSE claude-engine.jsClaudeOrg-level auth handling, SSE streaming, auto-recovery gemini-engine.jsGeminiSSE streaming with auto-reconnect perplexity-engine.jsPerplexitySSE streaming
CLI工具
这 proxima CLI允许您从终端使用任何AI提供程序。
安装
从应用程序
设置→ ⚡ 将CLI安装到PATH
来源
npm link # Windows
sudo npm link # macOS / Linux不安装
npm run cli -- ask "question"命令
# Ask any provider
proxima ask "How does async/await work in JS?"
proxima ask claude "Review this approach"
proxima ask chatgpt "Explain this error"
# Search
proxima search "latest Node.js release"
# Code
proxima code "REST API with Express and JWT auth"
proxima code review "function fetchUser(id) { ... }"
proxima code explain "async/await"
# Smart tools
proxima fix "SyntaxError: Unexpected token '&1 | proxima fix
# Review a git diff
git diff | proxima code review
# Pass file as context
proxima ask "What does this do?" --file src/server.js旗帜
FlagWhat it does -m / --modelOverride provider (claude, chatgpt, gemini, perplexity, auto) --jsonRaw JSON output for scripting -l / --langSpecify code language --fileInclude a file as context --toTarget language for translate --fromSource language for translate
______________________________________________________________________
REST API
Proxima在 http://localhost:3210.
在中启用它 设置→ REST API和CLI.
端点
POST /v1/chat/completions — OpenAI-compatible chat
GET /v1/models — List available models
GET /v1/functions — API function catalog with examples
GET /v1/stats — Response time stats per provider
POST /v1/conversations/new — Reset all conversations
GET /api/status — Server status
GET /docs — Interactive API docs (with live chat widget)
GET /cli — CLI documentation
GET /ws — WebSocket documentation函数
这 "function" 现场控制着发生的事情。无功能=正常聊天。
FunctionBody FieldsWhat it does (none)model, messageNormal chat "search"model, message, functionWeb search + AI analysis "translate"model, message, function, toTranslate text "brainstorm"model, message, functionGenerate ideas "code"model, message, function, actionCode generate/review/debug/explain "analyze"model, function, urlAnalyze URL or content "security_audit"model, code, functionScan code for vulnerabilities "debate"model, message, functionMulti-perspective debate
例子
聊天：
curl http://localhost:3210/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{"model": "claude", "message": "What is AI?"}'搜索：
curl http://localhost:3210/v1/chat/completions \
-d '{"model": "perplexity", "message": "AI news 2026", "function": "search"}'翻译：
curl http://localhost:3210/v1/chat/completions \
-d '{"model": "gemini", "message": "Hello world", "function": "translate", "to": "Hindi"}'代码生成：
curl http://localhost:3210/v1/chat/completions \
-d '{"model": "claude", "message": "Sort algorithm", "function": "code", "action": "generate", "language": "Python"}'查询所有提供商：
curl http://localhost:3210/v1/chat/completions \
-d '{"model": "all", "message": "Explain quantum computing"}'安全审计：
curl http://localhost:3210/v1/chat/completions \
-d '{"model": "claude", "function": "security_audit", "code": "db.query(\"SELECT * FROM users WHERE id=\" + req.query.id)"}'多模型查询
model: "all" // all enabled providers
model: ["claude", "chatgpt"] // specific providers响应格式
{
"id": "proxima-abc123",
"model": "claude",
"choices": [{
"message": {
"role": "assistant",
"content": "AI response here..."
}
}],
"proxima": {
"provider": "claude",
"responseTimeMs": 2400
}
}使用时 model: "all"，每个提供者在中都有自己的条目 choices[].
______________________________________________________________________
SDK
python
from proxima import Proxima
client = Proxima()
# Chat — any model
response = client.chat("Hello", model="claude")
response = client.chat("Hello", model="chatgpt")
response = client.chat("Hello") # auto picks best
print(response.text)
print(response.response_time_ms)
# Search
result = client.chat("AI news 2026",
model="perplexity", function="search")
# Translate
hindi = client.chat("Hello world",
model="gemini", function="translate",
to="Hindi")
# Code
code = client.chat("Sort algorithm",
model="claude", function="code",
action="generate", language="Python")
# System
models = client.get_models()
stats = client.get_stats()
client.new_conversation()pip install requests，然后复制 sdk/proxima.py 到你的项目。
JavaScript
const { Proxima } = require('./sdk/proxima');
const client = new Proxima();
// Chat — any model
const res = await client.chat("Hello",
{ model: "claude" });
console.log(res.text);
// Search
const news = await client.chat("AI news",
{ model: "perplexity",
function: "search" });
// Translate
const hindi = await client.chat("Hello",
{ model: "gemini",
function: "translate",
to: "Hindi" });
// Code generate
const code = await client.chat("Sort algo",
{ model: "claude",
function: "code",
action: "generate" });
// System
const models = await client.getModels();
const stats = await client.getStats();适用于Node.js 18+（原生） fetch).
SDK配置
client = Proxima(base_url="http://192.168.1.100:3210") # custom URL
client = Proxima(default_model="claude") # default model______________________________________________________________________
双向通信
实时流媒体AI ws://localhost:3210/ws.
要求在“设置”中启用REST API。
示例
const ws = new WebSocket("ws://localhost:3210/ws");
ws.send(JSON.stringify({
action: "ask",
model: "claude",
message: "What is a closure?",
id: "req_1"
}));
ws.onmessage = (e) => {
const msg = JSON.parse(e.data);
// { type: "status", id: "req_1", status: "processing", model: "claude" }
// { type: "response", id: "req_1", model: "claude", content: "...", responseTimeMs: 2400 }
};可用操作
ActionWhat it does ask / chatChat with any provider searchWeb search codegenerate / review / explain / optimize / debug translateTranslate text brainstormGenerate ideas debateMulti-provider debate (queries all providers) auditSecurity code audit new_conversationReset conversation context for all providers statsConnection and provider statistics pingKeepalive — returns pong
______________________________________________________________________
MCP工具
🤖 AI提供者工具
ToolWhat it does ask_chatgptQuery ChatGPT (supports file upload) ask_claudeQuery Claude (supports file upload) ask_geminiQuery Gemini (supports file upload) ask_perplexityQuery Perplexity (supports file upload) ask_all_aisSend same query to all providers at once ask_selectedPick specific providers to query compare_aisGet and compare responses side by side smart_queryAuto-picks best provider, falls back if one fails
🔧 开发工具
ToolWhat it does solveOne-shot problem solver — senior engineer level fix_errorRoot cause + exact fix for any error build_architectureFull project architecture blueprint write_testsGenerate tests (jest / vitest / mocha / pytest) explain_errorError explained in plain terms, no jargon convert_codeConvert code between languages or frameworks
⚔️ 多AI工具
ToolWhat it does chain_querySequential multi-AI pipeline — use {previous} to pass output forward debateMulti-provider debate with FOR / AGAINST / NEUTRAL stances verifyCross-provider answer verification with confidence score (0–100%) security_auditCode security scan — flags CRITICAL / HIGH / MEDIUM / LOW issues
💻 代码工具
ToolWhat it does generate_codeGenerate code from a description explain_codePlain-English explanation of any code optimize_codePerformance improvement suggestions review_codeCode review feedback verify_codeCheck against best practices
🔍 搜索工具
ToolWhat it does deep_searchComprehensive web search internet_searchGeneral internet search on any topic news_searchLatest news articles reddit_searchReddit discussions github_searchFind open-source repos, code, and solutions on GitHub academic_searchPapers and research math_searchMath problems step-by-step
📝 内容工具
ToolWhat it does brainstormGenerate ideas on any topic summarize_urlSummarize any URL generate_articleFull article generation writing_helpWriting assistance fact_checkFact verification find_statsFind statistics and data how_toStep-by-step instructions compareCompare two things in depth
🔬 分析工具
ToolWhat it does analyze_documentAnalyze documents from URL extract_dataExtract structured data from text or URL get_ui_referenceUI/UX design consultant — colors, layouts, components, CSS tokens, and code improvements
📁 文件工具
ToolWhat it does analyze_fileUpload and analyze a local file review_code_fileCode review on a local file (bugs, performance, security)
🪟 窗口控件
ToolWhat it does show_windowShow the Proxima window hide_windowHide to system tray toggle_windowToggle visibility set_headless_modeRun fully in background
🔄 会话工具
ToolWhat it does new_conversationReset conversation context clear_cacheClear response cache conversation_exportExport full conversation history
______________________________________________________________________
安全与隐私
由于Proxima在没有API密钥的情况下工作，因此有几点值得了解：
- 未存储凭据。 Proxima使用您现有的浏览器会话Cookie，与您已经登录的方式相同。
- 什么都不会离开你的机器 除了您向登录的AI提供商发送的查询。
- 在本地主机上运行。 MCP服务器、REST API和WebSocket都是本地的。没有任何东西暴露在互联网上。
- 没有遥测。 Proxima不会在任何地方收集或发送任何使用数据。
- 会议是你的。 如果您从提供商的网站注销或清除浏览器数据，则需要通过Proxima重新登录。
Proxima不会绕过身份验证，它使用您已有的会话。与在浏览器中使用该网站相同。
______________________________________________________________________
项目结构
Proxima/
├── electron/
│ ├── main-v2.cjs # Electron main process
│ ├── browser-manager.cjs # Browser session management
│ ├── rest-api.cjs # REST API server (OpenAI-compatible)
│ ├── ws-server.cjs # WebSocket server
│ ├── provider-api.cjs # Provider engine injection manager
│ ├── index-v2.html # App UI
│ ├── preload.cjs # Renderer preload bridge
│ └── providers/
│ ├── chatgpt-engine.js # SHA3-512 POW + SSE streaming
│ ├── claude-engine.js # Org auth + SSE streaming
│ ├── gemini-engine.js # Session SSE streaming
│ └── perplexity-engine.js # SSE streaming
├── cli/
│ └── proxima-cli.cjs # Terminal CLI
├── src/
│ ├── mcp-server-v3.js # MCP server (50+ tools)
│ └── enabled-providers.json # Provider config
├── sdk/
│ ├── proxima.py # Python SDK
│ └── proxima.js # JavaScript SDK
├── assets/ # Icons, screenshots, demo
└── package.json______________________________________________________________________
故障排除
首次启动时的Windows防火墙提示
Proxima继续运行 localhost:19223 和 localhost:3210。单击“允许”--它只接受本地连接。
提供商显示“未登录”
每个提供者都有不同的登录方法：
- 克劳德·查特GPT，困惑 --单击提供者选项卡并使用OTP（电子邮件代码）登录。根据谷歌的政策，谷歌登录在嵌入式浏览器中受到限制。
- 双子座 --使用基于cookie的身份验证。首先在您的常规浏览器中登录谷歌，然后Proxima会自动接听会话。
REST API未响应
检查是否在“设置”中启用了REST API→ REST API和CLI部分。访问 http://localhost:3210 在浏览器中进行验证。
MCP工具未在编辑器中显示
- 确保Proxima正在运行
- 验证MCP配置中的路径（使用设置复制按钮）
- 重新启动编辑器
CLI： proxima 安装后找不到
打开一个新的终端。如果仍未找到，请单击 🔧 修复 在设置中→ CLI部分。
CLI：“无法连接到Proxima”
Proxima必须正在运行，并且必须启用REST API。CLI连接到 localhost:3210.
WebSocket无法连接
WebSocket共享REST API服务器。首先在“设置”中启用REST API。
______________________________________________________________________
赞助商💖
随着Proxima的不断发展， ](https://github.com/sponsors/Zen4-bit) 有助于维护和改进项目
许可证
Proxima已获得许可 非商业使用软件。参见 许可证 完整条款。
______________________________________________________________________
Proxima v4.1.0 -一个API，所有人工智能模型⚡
由...制作 Zen4位 ·每一个⭐ 事情💕