Skip to content

README_EN.md

Latest commit

 

History

History
234 lines (175 loc) · 9.59 KB

README_EN.md

File metadata and controls

234 lines (175 loc) · 9.59 KB

Antigravity Tools (2API Edition) 🚀

Antigravity Logo

Your Personal API Gateway for Infinite AI

Seamlessly proxy Gemini & Claude. OpenAI-Compatible. Privacy First.

Version Tauri React License

📥 Download • ✨ Account Manager • 🔌 API Proxy

🇺🇸 English | 🇨🇳 简体中文 (Legacy v2)

Antigravity Tools 2.2 is a robust desktop application that transforms your desktop into a powerful Local AI Gateway.

It not only manages your Gemini / Claude accounts but also provides a local OpenAI-compatible API server. This allows you to use your browser-based Google/Claude sessions (sid, __Secure-1PSID, etc.) as standard API keys in any AI application (Cursor, Windsurf, LangChain, etc.).

Looking for the Account Manager Only version? The v2.0 Account Manager documentation has been moved to README_v2.md.

🔌 API Proxy: In-Depth

Antigravity's proxy is not just a forwarder, but a complete Local AI Scheduler.

1. 🔄 Smart Account Rotation

When you have multiple accounts:

  • Load Balancing: Requests are distributed across healthy accounts.
  • Auto Failover: If an account hits 429 (Rate Limit), the system instantly switches to the next account and retries.
  • Quota Awareness: Exhausted accounts are automatically skipped.

2. 🧠 Perfect Context

Fully compatible with OpenAI's messages format. Multi-turn conversations work seamlessly in apps like Cursor, Windsurf, or NextChat.

3. 🔗 Multi-Protocol Support (NEW!)

Supports both OpenAI and Anthropic API formats:

  • OpenAI Protocol: /v1/chat/completions - Compatible with Cursor, Windsurf, NextChat, etc.
  • Anthropic Protocol: /v1/messages - Native support for Claude Code CLI and other tools
  • Auto Conversion: Automatically converts requests to Gemini format for perfect compatibility
📘 Claude Code CLI Setup Guide

Quick Setup

Temporary Use (run each time you open terminal):

export ANTHROPIC_API_KEY="sk-antigravity"  # Use the API Key from the app
export ANTHROPIC_BASE_URL="http://127.0.0.1:8045"
claude "write a quicksort algorithm"

Permanent Setup (one-time configuration):

# Edit config file (macOS/Linux)
nano ~/.zshrc

# Add these lines to the end
export ANTHROPIC_API_KEY="sk-antigravity"
export ANTHROPIC_BASE_URL="http://127.0.0.1:8045"

# Reload config
source ~/.zshrc

Usage Examples

# Basic conversation
claude "what is 1+1"

# Multi-turn conversation
claude "what did I just ask"

# Code generation
claude "write a quicksort in Python"

Notes

  1. Ensure Service Running: Make sure Antigravity API Proxy is running before use
  2. API Key: Copy the generated API Key from "API Proxy" page
  3. Port Configuration: Default port is 8045, update ANTHROPIC_BASE_URL if changed

3. Connect

Gemini 3 Pro Image (Imagen 3) Claude 3.5 Sonnet (Thinking)
NextChat - Image Gen/Edit Windsurf/Cursor - Complex Coding

👥 Account Manager

  • Token Management: Manage dozens of Gemini/Claude accounts.
  • Auto-Refresh: Keeps your tokens alive automatically.
  • Quota Monitoring: Real-time visualization of model quotas (Text & Image).
  • Account Switching: One-click token injection into local Antigravity database for seamless switching.

🛡️ Privacy First

  • Local Storage: All data inside gui_config.json and antigravity.db stays on your machine.
  • No Cloud: We do not run any intermediary servers. Your data goes directly from your machine to Google/Anthropic.

🛠️ Technology Stack

Component Tech
Core Rust (Tauri v2)
API Server Axum (Rust)
Frontend React + TailwindCSS
Database SQLite + JSON

📦 Usage

  1. Add Accounts: Login via OAuth or paste tokens in the "Accounts" tab.
  2. Start Proxy: Go to "API Proxy" tab and click Start Service.
  3. Connect:
    • Base URL: http://localhost:8045/ (Some apps need http://localhost:8045/v1)
    • API Key: sk-antigravity (Any string)
    • Model: Select from the list below:

📚 Supported Models

Model ID Description
gemini-2.5-flash Flash 2.5. Extremely fast and cost-effective.
gemini-2.5-flash-thinking Flash Thinking. Lightweight model with reasoning capabilities.
gemini-3-pro-high Gemini 3 Pro. Google's strongest reasoning model.
gemini-3-pro-low Gemini 3 Pro (Low). Lower quota consumption version.
gemini-3-pro-image Imagen 3. Dedicated image generation model.
claude-sonnet-4-5 Claude 3.5 Sonnet. Top choice for coding and logic.
claude-sonnet-4-5-thinking Sonnet Thinking. Sonnet with chain-of-thought enabled.
claude-opus-4-5-thinking Opus Thinking. Claude's most powerful thinking model.

💡 Tip: The proxy supports pass-through for all official Google/Anthropic model IDs.

🔄 Changelog

v3.1.1 (2025-12-18)

🌐 Global Upstream Proxy Support

  • System-wide Proxy Integration: Added support for global upstream proxies (HTTP/SOCKS5). When enabled, all outgoing requests (API proxy, token refresh, quota checks, update checks) are routed through the configured proxy.
  • One-Click Hot-Reloading:
    • Proxy settings and model mappings take effect instantly upon saving.
    • The running proxy server automatically reloads configuration without requiring a restart.
  • Backend Architecture Optimization:
    • Centralized HTTP client management for improved connection stability.
    • Thread-safe dynamic configuration sharing via Arc<RwLock>.
  • UI Improvements: Added a dedicated "Proxy" tab in Settings for better configuration management.

🧹 Code Quality & Cleanup

  • Fixed compiler warnings regarding unused fields in server.rs.
  • Removed redundant code and outdated imports to improve performance.

v3.1.0 (2025-12-18)

🔧 API Proxy Optimizations

  • 403 Error Smart Handling: Instantly identifies and marks accounts with 403 Forbidden, no more retries
    • Auto-marks as "forbidden" status
    • Auto-skips 403 accounts during batch refresh
    • Saves 3+ seconds response time

v3.0.2 (2025-12-17)

🔧 API Proxy Optimizations

  • 403 Error Smart Handling: Instantly identifies and marks accounts with 403 Forbidden, no more retries

    • Auto-marks as "forbidden" status
    • Auto-skips 403 accounts during batch refresh
    • Saves 3+ seconds response time

  • Claude CLI Response Optimization: Fixed empty response and JSON format issues

    • Increased maxOutputTokens from 8096 to 16384 for longer responses
    • Removed toolConfig to avoid MALFORMED_FUNCTION_CALL errors
    • Added detailed diagnostic logs recording raw Gemini responses

  • Logging System Enhancement:

    • Records full candidates data for empty text responses
    • Increased log display length from 60 to 100 characters
    • Distinguished log levels for empty vs normal responses

🐛 Bug Fixes

  • OAuth Environment Check Optimization: Simplified Tauri environment check, only validates invoke function availability
    • Removed window.__TAURI__ check
    • Avoids false positives in certain Tauri versions

v3.0.1 (2025-12-17)

🔧 Bug Fixes

  • macOS Process Termination Refactor (Critical): Completely resolved the "Unexpected Termination" dialog issue during account switching on macOS. We rewrote the process detection algorithm to intelligently identify the main process based on process arguments and characteristics (filtering out Helpers), enabling a 100% graceful exit via targeted SIGTERM, while maintaining a safety fallback for force cleanup.
  • Image Generation Optimization: Added support for gemini-3-pro-image and various aspect ratio suffixes (e.g., -1:1, -16:9). New models: gemini-3-pro-image-4x3, gemini-3-pro-image-3x4, gemini-3-pro-image-4k, gemini-3-pro-image-16x9-4k
    • Parameter support: size now accepts 1024x768 (4:3) and 768x1024 (3:4)
    • 4K HD support: via -4k suffix or "quality": "hd" parameter

v3.0.0 (2025-12-16)

  • 🚀 Initial API Proxy release
  • 🔌 Built-in high-performance Rust proxy server
  • 🔄 Smart account rotation and auto-failover
  • 🧠 Full OpenAI protocol compatibility
  • 🖼️ Gemini Imagen 3 image generation support

📄 License

CC BY-NC-SA 4.0