v1.4.0

🚀 Major Release: v1.4.0

This is a massive feature release that brings a brand-new Web proxy engine, an advanced CLI/IDE account switcher, UI dashboard improvements, and strict Antigravity client sync.

🔥 Major New Features

• Gemini Web Provider (gemini_web): A completely new authentication and routing engine that connects to your local Chrome browser via CDP (port 9222) to automatically extract Google session cookies and SNlM0e CSRF tokens. This allows agy-proxy to seamlessly proxy requests directly to the free gemini.google.com backend using high-performance HTTP polling without requiring API keys!
• Universal Session Switcher: A brand new interactive CLI and Web Dashboard feature to effortlessly switch the active Google Antigravity account.
  • Switch accounts directly from the terminal using agy-proxy switch "email@gmail.com".
  • Supports separate targets using --cli (terminal) and --ide (Cursor/VSCode) flags.
  • Automatically writes the strict 3-key json.Marshal schema expected by the official Go client.
  • Zero-config auto-discovery of all existing local tokens.
• Advanced Dashboard UI:
  • Sectioned grouping for OAuth, Gemini Web, and API Keys.
  • 1-click collapse/expand for clean management of dozens of accounts.
  • Live CLI vs IDE status indicators directly on the account cards.
  • Sandbox guardrails and real-time streaming RPC parsing.

🐛 Bug Fixes & Stability

• Exact Schema Generation: Fixed token serialization to perfectly match Antigravity's expected binary layout, preventing CLI crashes.
• PermissionError Fix: allow_overwrite=True now correctly bypasses the AGY_ALLOW_CLI_TOKEN_OVERWRITE env guard, ensuring the UI and CLI can always switch accounts smoothly.
• Misleading Status Fix: Replaced the confusing "Active (Primary)" status with accurate filesystem-level "Active (CLI)" and "Active (IDE)" indicators in the agy-proxy switch --list output.
• Stream Parsing: Resolved HTTP 400 errors in StreamGenerate parsing for web connections.
• Client Protocol Sync: Synced Antigravity 1.2.7 client protocols and updated upstream User-Agent to antigravity/cli/1.2.7 (cl=984112147) to bypass strict backend detection.

🛡️ Security

• Atomic Writes & Backups: Added automatic .bak backups before modifying any antigravity-oauth-token file.
• Protected Host Tokens: Hardened UI controls to prevent accidental overwrites of system-level tokens.

📋 Full Changelog

(Includes all commits since v1.3.0)

• 3781ea5 chore: bump version to 1.4.0 (major feature update)
• 4dcfda5 chore: bump version to 1.3.1
• 7e315c3 feat(auth): update upstream User-Agent to antigravity/cli/1.2.7
• bc126e9 fix(cli): replace misleading Primary status with actual CLI/IDE active status
• cf425a4 fix(switcher): allow_overwrite=True bypasses AGY_ALLOW_CLI_TOKEN_OVERWRITE
• 5f30691 feat(gemini_web): synthesize tool calls from web responses
• eb25955 fix(gemini_web): resolve StreamGenerate HTTP 400 and stream parsing
• e0b45ae feat(web): anti-sandbox guardrails, live Set-Cookie rotation
• 7b2bb51 feat(web): reverse-engineer StreamGenerate protocol
• 5611ddb feat: sectioned accounts view, multi-account web/apikey support
• b47778c feat(web): add Gemini Web HAR/cookie import
• 93c2c70 feat(security): add backups, atomic writes, readonly mode
• 1a4ace5 feat: add Antigravity session switcher CLI, API and UI rotation
• 65719a4 feat(auth): add GeminiWebSession with CDP auto-cookie extraction
• 6ab5421 feat(cloudcode): add universal v1internal passthrough router

Full Changelog: v1.3.0...v1.4.0

---

All tests (198/198) pass successfully.

v1.3.0

Antigravity Proxy v1.3.0 — Gemini 3.8 Flash, Modular 5-Tab Gateway, Smart Tool Pruning & Dual Quota Analytics

We are excited to release Antigravity Proxy v1.3.0, a major feature release bringing first-class support for Google's newest Gemini 3.8 Flash model family, an overhaul of the Web Dashboard into a modular 5-tab gateway, proactive Smart Tool Pruning, and accurate Dual Quota Bucket Analytics.

---

🌟 Highlights & Key Features

1. ⚡ Gemini 3.8 Flash Family & Smart Model Routing

• New Default Model: Upgraded default model to gemini-3.8-flash-high for lightning-fast responses with deep reasoning capabilities.
• Full Gemini 3.8 Lineup: Added support for gemini-3.8-flash-high, gemini-3.8-flash-low, and gemini-3.8-flash-tiered.
• Deprecation Fallback: Automatically redirects legacy gemini-3-flash-agent calls to gemini-3.8-flash-tiered.
• Agent Request Classification: Automatically classifies and tags tool/function-calling requests with upstream agent request types.
• Clean /v1/models Catalog: Advertises only verified, authentic models to client applications.

2. 🤖 Enhanced Claude Code & claude-cli Integration

• Native Handshake Endpoints: Added Anthropic and Claude OAuth hello endpoints to provide seamless drop-in compatibility with claude-cli.
• Dynamic Model Discovery: Claude launcher now dynamically queries available models from the proxy rather than relying on a hardcoded list.
• Anthropic Aliases: Full namespace support for anthropic.* model aliases.

3. 🧠 Smart Tool Pruning & Context Compactor v2

• Smart Tool Pruning: Proactively truncates bulky historical tool outputs in long-running agentic conversations, dramatically reducing context bloat and avoiding token limits.
• Auto-Compactor v2: Upgraded context compaction engine with extended memory windows, rich previews, and adaptive thinking support. Dedicated gemini-3.8-flash-low to compaction duties while keeping gemini-3.1-flash-lite for lightweight background tasks.
• Extended Capabilities: Added native token counting, Google Trawler URL fetcher, and Structured Outputs schema parsing.

4. 📊 Dual Quota Analytics (5h & Weekly Windows)

• Dual Quota Breakdown: Full visibility into both the 5-hour burst window and 7-day rolling quota windows via v1internal:retrieveUserQuotaSummary.
• Periodic Background Sync: Automatic quota synchronization in the background without adding latency to chat requests.
• Accurate Metrics & Persistence: Fixed bottleneck calculations for Gemini and 3P (Claude/GPT), accurate rate-limit states, and persistent request counters across pool reloads.

5. 🎨 Brand New Modular 5-Tab Dashboard UI

• 5 Dedicated Tabs:
  • Accounts: Multi-account pool health, OAuth management, and priority switches.
  • Analytics: Real-time dual quota visualizers, 3P KPI cards, and usage graphs.
  • Playground: Redesigned testing arena with foldable "Thinking" accordions, KaTeX/math equation rendering, and robust clipboard copying.
  • Integrations: Quick-start configuration guides for Claude Code, Cursor, Cline, LibreChat, and OpenAI clients.
  • Settings: Proxy-wide cache, compactor, and security configurations.
• Responsive Layout: Optimized for both mobile devices and high-resolution desktop screens with zero header button clipping.

6. 🔒 Security & Upstream Protocol Sync

• Isolated Configuration: Account persistence is strictly isolated to ~/.config/agy-proxy/accounts.json, ensuring native Antigravity CLI/IDE token files are never modified or overwritten.
• 401 Auto-Failover: Transparent force-refresh and seamless rotation across pool accounts upon receiving upstream unauthorized errors.
• Upstream Sync: Synchronized upstream User-Agent protocol to antigravity/cli/1.1.28 (cl=978129418) across both the proxy server and Cloudflare Worker.

---

📦 Quick Install & Upgrade

Via pip / uv:

pip install --upgrade agy-proxy
# or using uv:
uv tool upgrade agy-proxy

Self-Update Command:

agy-proxy update

1-Click Install Script (Linux / macOS):

curl -fsSL https://raw.githubusercontent.com/kqlio67/agy-proxy/main/install.sh | bash

1-Click Install Script (Windows PowerShell):

irm https://raw.githubusercontent.com/kqlio67/agy-proxy/main/install.ps1 | iex

Docker / GHCR:

docker run -d -p 8045:8045 -v ~/.config/agy-proxy:/root/.config/agy-proxy ghcr.io/kqlio67/agy-proxy:latest

---

Full Changelog: v1.2.0...v1.3.0

v1.2.0

⚡ Antigravity Proxy v1.2.0 — Context Compactor, Live Google Search & Windows 1-Click Installers

We are excited to release Antigravity Proxy v1.2.0! This major update introduces our intelligent Context Compactor Engine, official Claude Code /compact interception, native Google Grounding live web search, 1-click installers for Windows, and comprehensive standalone package builds for all major platforms.

---

🌟 What's New in v1.2.0

🗜 1. Context Compactor & Auto-Summarization Engine

• 80% to 98% Token Reduction: Automatically monitors session token counts and compresses large conversation histories down into a clean ~1,500-token structured context, allowing coding sessions to run indefinitely without hitting Google's hard limit of 1,048,576 tokens (Error 400 Context Overflow).
• Official Claude Code /compact Interception: Intercepts Claude Code's manual /compact and /autocompact commands, routing summarization to gemini-3.1-flash-lite using the official 9-section summarization schema while preserving 100% of your primary model quota.
• Dynamic Dashboard Controls: Interactive threshold slider (30k–200k tokens) and configurable uncompressed message preservation (Preserve Last Messages: 4–16 msgs) directly from the Web UI.

🌐 2. Live Web Search & Google Grounding (Vertex AI)

• Native Google Search: Empowers coding assistants (like Claude Code) with real-time web search capabilities via Google Grounding without requiring paid third-party search API keys.
• Concurrent Multi-Engine Fallback: Automatically queries DuckDuckGo, Bing, and Brave in parallel with domain filtering (site:), URL deduplication, and citation generation.

⚡ 3. Intelligent Background Task Optimization

• Quota Preservation: Automatically detects non-interactive background tasks (title generation, context compaction, checkpoints) and transparently routes them to gemini-3.1-flash-lite with a 0-thinking budget, saving your primary reasoning model quota.

🛡 4. Zero-Delay Recovery (Retry-After: 2)

• Prevents Claude Code from locking up in 4-minute exponential backoff retry loops during transient upstream errors by injecting standard Retry-After: 2 headers for instantaneous 2-second failovers across pool accounts.

💻 5. One-Click Installers & Cross-Platform Binaries

• Added instant 1-click installer scripts for Windows PowerShell (install.ps1) and Windows Command Prompt (install.cmd), automatically setting up the executable in the user's system PATH.
• Full matrix build support: Windows x64, Linux x64/ARM64, macOS ARM64 (Apple Silicon), and macOS Intel (x86_64).
• Standalone Python Wheels (.whl) and Source Distributions (.tar.gz) published alongside every release.

🧹 6. Minimalist Logging & Protocol 1.1.23

• Updated User-Agent to the official Antigravity CLI 1.1.23 protocol (cl=974125021).
• Stripped console clutter and emojis from standard user logs; refined the --debug flag for full technical developer tracing.
• Updated Cloudflare Edge Worker (cloudflare/worker.js) to support the 1.1.23 protocol and fast error recovery.

---

📦 Download Options

⚡ 1-Click Fast Install (No Python Required)

🔹 Linux & macOS (Bash):

curl -fsSL https://raw.githubusercontent.com/kqlio67/agy-proxy/main/install.sh | bash

🔹 Windows (PowerShell):

irm https://raw.githubusercontent.com/kqlio67/agy-proxy/main/install.ps1 | iex

🔹 Windows (Command Prompt / CMD):

curl -fsSL https://raw.githubusercontent.com/kqlio67/agy-proxy/main/install.cmd -o install.cmd && install.cmd && del install.cmd

---

🖥 Standalone Executables (Pre-built Binaries)

Platform	Architecture	Archive
Windows	x64 (amd64)	agy-proxy-windows-amd64.zip
Linux	x64 (amd64)	agy-proxy-linux-amd64.tar.gz
Linux ARM64	ARM64 (Raspberry Pi, VPS)	agy-proxy-linux-arm64.tar.gz
macOS	Apple Silicon (M1/M2/M3/M4)	agy-proxy-darwin-arm64.tar.gz
macOS	Intel (x86_64)	agy-proxy-darwin-amd64.tar.gz

---

🐍 Python Package (PyPI / Pip)

pip install agy-proxy==1.2.0

• Python Wheel: agy_proxy-1.2.0-py3-none-any.whl
• Source Archive: agy_proxy-1.2.0.tar.gz

---

🐳 Docker Container (GHCR)

docker pull ghcr.io/kqlio67/agy-proxy:v1.2.0
docker pull ghcr.io/kqlio67/agy-proxy:latest

---

🔒 Verification

All release assets include verifiable cryptographic SHA256 hashes inside SHA256SUMS.txt.

sha256sum -c SHA256SUMS.txt

---

📋 Full Commits Changelog

• e999396 chore(cli): refine clear separation between standard minimalist user mode and full --debug developer mode
• 3fb77a6 feat(installer): add 1-click Windows PowerShell (install.ps1) and CMD (install.cmd) installers
• f2bb168 chore(release): bump version to v1.2.0 and add python wheels, intel mac, sha256 checksums to release workflow
• dab1492 fix(cloudflare): update worker User-Agent to 1.1.23 protocol and add Retry-After header
• 6be8944 docs(readme): add comprehensive documentation for Context Compactor, Google Grounding search, and background routing
• 0cc91a5 chore(logging): strip all remaining emojis from logs, banners, and compactor status
• 35dc89f chore(logging): remove lightning emoji from request and cache logs
• 896acc8 perf(cache): memoize unsupported Free Tier API keys to prevent redundant cachedContents 429 calls
• 972fa60 refactor(compactor): make estimate_message_tokens fully unified, null-safe, and comprehensive
• d159c70 fix(openai): initialize model and req_id in stream_openai_chat
• a43719d chore(logging): silence uvicorn polling access logs and httpx spam in console
• f2e0c8e feat: add context auto-compactor engine, official Claude /compact interception, and UI controls
• e661bac feat(core): Antigravity CLI 1.1.23 integration, background model optimization, native Google Search & multi-account stability
• 0ae0228 feat(search): add built-in live WebSearch engine and launcher danger mode prompt

v1.1.0

⚡ Antigravity Proxy v1.1.0 — Prompt Caching, UI Redesign, 1-Click Installers & Auto-Updater

We are excited to announce Antigravity Proxy v1.1.0! 🎉
This major update brings intelligent Prompt Caching & Session Affinity (saving up to 75% on input tokens), a completely redesigned Web Dashboard with Compact Table & Search, 1-Click Installers, native Docker/GHCR support, a built-in CLI Self-Updater, and comprehensive stability fixes.

---

🚀 Key Highlights & New Features

1. 🧠 Intelligent Prompt Caching & Session Affinity (75% Token Discount)

• Session Affinity ("Sticky Sessions"): Deterministically pins multi-turn conversations (Claude Code, Cursor, Windsurf) to stable backend accounts, maximizing Google's internal KV-cache reuse.
• Google Native Context Caching: Integrates Google AI Studio's cachedContents API for large prompts (>30k chars), eliminating duplicate system instructions and tool definitions with a 75% token discount and near-zero first-token latency.
• Live Cache Metrics: Real-time token savings counter and stats exposed via /api/cache/stats and displayed in the Web UI (Cache: ~X.Xk saved).

2. 📊 Web Dashboard Redesign & Multi-Account Pool Scaling

• Compact Table View 📄: Toggle between Grid and Compact Table views to easily monitor 20+ accounts on a single screen without scrolling.
• Real-Time Search & Category Filters: Instantly filter accounts by name, email, API key prefix, or status (All, OAuth, API Keys, 0% Exhausted).
• Live Active Model Tracker: Real-time display of the latest executing model and API source directly inside account cards and table rows.
• 1-Click Integration Snippets: Redesigned Client Integrations section with instant copy-paste setup guides for Claude Code CLI, Cursor, Windsurf, Roo Code / Cline, Continue.dev, Aider, Python SDK, and Cloudflare.

3. 📦 1-Click Installation, Standalone Binaries & Docker

• 1-Line Shell Installer (Linux & macOS):

  curl -fsSL https://raw.githubusercontent.com/kqlio67/agy-proxy/main/install.sh | bash

• Official Container Image (GHCR):

  docker run -d -p 8000:8000 \
    -v ~/.config/agy-proxy:/home/appuser/.config/agy-proxy \
    --name agy-proxy ghcr.io/kqlio67/agy-proxy:latest

• Zero-Dependency Standalone Binaries: Automated pre-compiled single binaries attached for Windows x64 (.exe), macOS Apple Silicon (M1–M4), Linux x86_64, and Linux ARM64 (Raspberry Pi/Orange Pi).

4. 🔄 In-Place CLI Self-Updater (agy-proxy update)

• Check and upgrade Antigravity Proxy directly from your terminal:

  # Check for updates:
  agy-proxy update --check
  
  # Automatically download & apply the latest release:
  agy-proxy update

5. 🛡️ Enhanced Failover, Error Handling & Clean Logs

• Model Overload & Deprecation Protection: Transparent routing for deprecated Gemini models (e.g. 2.5-flash $\rightarrow$ 3.7-flash / 3.6-flash), and graceful handling of 503 Service Unavailable without unauthorized model substitution.
• Graceful Shutdown: Clean Ctrl+C handling without ASGI / starlette traceback spam.
• Ultra-Clean Production Logging: Silenced internal HTTP client noise (httpx/uvicorn.access) by default for single-line readable logs ([Account] ⚡ Model).
• 100% GitHub Community Standards: Added official Code of Conduct, Contributing Guide, Security Policy, and interactive Issue/PR Forms.

---

🧪 Verification

• 100% Test Suite Pass: Verified across all 11 integration suites including multi-turn Anthropic tool-use loops, streaming SSE, and simulated quota failovers.

---

📥 Quick Install & Upgrade

# Upgrade existing installation:
agy-proxy update

# Or install fresh (Linux/macOS):
curl -fsSL https://raw.githubusercontent.com/kqlio67/agy-proxy/main/install.sh | bash

# Or via Python / UV:
git pull origin main
uv run python main.py --port 8000

v1.0.2

⚡ Antigravity Proxy v1.0.2 Release Notes

🚀 What's New in v1.0.2

1. 🤖 Deep & Optimized Claude Code CLI Integration

• CLI Argument Passthrough: All native Claude Code commands and flags (-c, -r, -p, --dangerously-skip-permissions, doctor, plugins, agents, etc.) are now seamlessly passed through the launcher scripts (./run_claude.sh for Linux/macOS and run_claude.bat for Windows).
• Sub-millisecond Service Stubs: Added local in-memory endpoints for /api/claude_cli/bootstrap, /api/claude_code_penguin_mode, /api/web/domain_info, and /api/oauth/claude_cli/create_api_key, reducing session startup and initialization time to < 1 ms.
• Security Classifier Auto-Routing: The background auto-mode security monitor (claude-sonnet-5) is now transparently routed to gemini-3.7-flash-high, completely eliminating auto-mode timeouts and command execution pauses.
• Model Picker Support (/model): All Antigravity backend models are now available and selectable in the interactive /model picker without whitelist restrictions.

2. 🌟 Expanded Model Aliases (Claude 5 Family)

• Added native routing support for the latest model generation:
  • claude-fable-5, claude-fable-5-latest, claude-5-fable $\rightarrow$ gemini-3.7-flash-high
  • claude-opus-5, claude-opus-5-latest, claude-5-opus $\rightarrow$ claude-opus-4-6-thinking
  • claude-haiku-4-5, claude-haiku-4-5-20251001, claude-4-5-haiku $\rightarrow$ gemini-3.1-flash-lite

3. 🛡️ Enhanced Stability & Multi-Account Failover

• Expanded Failover Coverage: Automatic pool rotation now handles 403 Forbidden, 500/502/503/504 Service Unavailable, and network timeouts (httpx.TimeoutException, httpx.NetworkError) in addition to 429 Too Many Requests.
• Concurrency Protection: Added asyncio.Lock() per account session for thread-safe OAuth token refreshes during high-concurrency parallel tool calls.
• Message & Schema Sanitization: Automatic filtering of empty content blocks ({"text": ""}) and strict JSON Schema translation for robust long-running Tool Calling chains.

4. 🔑 Streamlined CLI Authentication & Usability

• Dedicated Auth Subcommands:
  • agy-proxy auth login — Google Antigravity browser OAuth (PKCE).
  • agy-proxy auth api (or apikey) — Add Google AI Studio Gemini API Keys (--key / --name).
  • agy-proxy auth list — Clean table view of pooled accounts, active projects, and live quotas.
• Graceful Exit: Handled Ctrl+C (KeyboardInterrupt / EOFError) across all CLI commands for clean cancellations without Python tracebacks.

5. 📊 Web Dashboard Updates (Client Integrations)

• Updated the Client Integrations section with instant copy-paste setup guides for:
  • Claude Code CLI
  • Cursor / Windsurf
  • VS Code (Roo Code / Cline / Continue.dev)
  • Aider CLI
  • Python SDK (OpenAI & Anthropic)
  • Gemini Native API (v1beta)
  • Cloudflare Tunnel / Edge Worker

---

🧪 Testing & Verification

• 11 out of 11 end-to-end integration tests passing with 100% success (uv run python test_proxy.py).
• Verified multi-turn tool calling across 10+ consecutive execution steps and simulated quota exhaustion (429 Failover).

v1.0.1

Release v1.0.1 — Claude Code Background Token Protection & Model Routing Fix

Overview

Release v1.0.1 resolves unintended Claude Sonnet token consumption caused by Claude Code CLI's internal background subroutines, optimizes lightweight model aliases, and ensures seamless explicit model selection across all supported AI providers.

---

Key Highlights & Changes

1. 🛡️ Claude Code Auxiliary Token Protection

• Problem: When using Claude Code CLI, internal auxiliary operations (such as conversation compaction, summary generation, tool result compression, and bash command safety checks) default to sending requests with the model identifier claude-sonnet-5. Previously, generic keyword matching routed these background requests to the native claude-sonnet-4-6 backend, prematurely exhausting Sonnet quotas.
• Solution: Explicit aliases for claude-sonnet-5, claude-sonnet-5-latest, and claude-5-sonnet now automatically map to gemini-3.7-flash-high. Background maintenance tasks run instantly and cost-effectively without touching valuable Claude Sonnet quotas.

2. ⚡ Lightweight Haiku Model Mapping

• Added aliases for claude-3-haiku and claude-3-5-haiku to automatically route to gemini-3.1-flash-lite for high-speed, lightweight task execution.

3. 🎯 Preserved Native Model Routing

• Explicit user requests remain 100% direct and unmodified:
  • claude-sonnet-4-6, claude-3-7-sonnet, claude-3-5-sonnet $\rightarrow$ Native Claude Sonnet
  • claude-opus-4-6-thinking, claude-3-opus $\rightarrow$ Native Claude Opus Thinking
  • gemini-3.7-flash-high, gemini-2.5-pro, gemini-3.1-pro-high $\rightarrow$ Native Google Gemini

---

Validation & Testing

• Unit & Normalization Tests: Passed 100% across all model permutations and aliases.
• Integration Test Suite: All 10 end-to-end integration tests passed, verifying:
  • Web Dashboard and Account Pool API lifecycle
  • OpenAI Chat Completions (Non-Streaming & SSE Streaming)
  • Anthropic Claude Messages (Non-Streaming & SSE Streaming with Thinking blocks)
  • Multi-turn Tool Calling & Thought Signature echoing
  • Token Counting endpoint (/v1/messages/count_tokens)

---

Full Changelog: v1.0.0...v1.0.1

v1.0.0

⚡ Antigravity Proxy v1.0.0 — Initial Release

We are excited to announce the first official release of Antigravity Proxy! 🚀

A high-performance, lightweight Python & standalone proxy server that exposes Google Antigravity & Gemini Code Assist as standard OpenAI (/v1/chat/completions), Anthropic Claude (/v1/messages), and Gemini Native (/v1beta/...) API endpoints.

---

🌟 Key Highlights

👥 Multi-Account Pooling & Limit Bypass

• Multi-Account Pooling: Pool multiple Google accounts simultaneously to multiply your rate limits and concurrent throughput.
• Automatic 429 Failover: When Account A exhausts its quota bucket, requests seamlessly route to Account B without dropping your session!
• Least-Used Load Balancing: Evenly distributes parallel calls across healthy accounts.

🧠 Gemini 3.7 Thinking & Protocol Translation

• Thinking Process Extraction: Preserves reasoning tokens into delta.reasoning_content (OpenAI format) and thinking blocks (Anthropic Claude format).
• Thought Signature Preservation: Multi-turn tool calling maintains thoughtSignature across turns to ensure compatibility with Claude Code, Cursor, and Aider.
• Multimodal Support: Full support for base64 image URIs and public image URLs.

📊 Modern Web Dashboard & Live Playground

• Real-time Quota Gauges: Monitor Gemini 5-hour and Claude quota bars per account.
• Account Manager: Add Google accounts via OAuth PKCE or Google AI Studio API keys, toggle/pause accounts, or remove them with one click.
• Interactive Playground: Test models with markdown rendering and collapsible thinking bubbles.

📦 Pre-built Standalone Binaries (Zero Dependencies)

Download the single executable for your OS from the Assets section below:

• 🪟 Windows (agy-proxy.exe): agy-proxy-windows-amd64.zip
• 🐧 Linux (x86_64): agy-proxy-linux-amd64.tar.gz
• 🍓 Linux ARM64 (Raspberry Pi 3/4/5 / ARM VPS): agy-proxy-linux-arm64.tar.gz
• 🍏 macOS (Apple Silicon M1/M2/M3/M4): agy-proxy-darwin-arm64.tar.gz

🌐 Cloudflare Edge Worker (Dual-Mode)

• Mode 1 (Geo-Bypass): Route local proxy traffic through Cloudflare Edge to bypass regional IP blocks with 0% VPN overhead (--cloudflare-url).
• Mode 2 (100% Serverless): Run the proxy 24/7 in the cloud with zero local PC running (npx wrangler secret put ACCOUNTS_JSON).

📱 Cross-Platform & 1-Click Launchers

• Windows: start_proxy.bat and run_claude.bat
• Linux & macOS: ./start_proxy.sh and ./run_claude.sh
• Android: Full native Termux support (pkg install python git -> python main.py).

---

🚀 Quick Start

Option 1: Standalone Binary (No Python required)

Extract the downloaded binary and run:

# Linux / macOS:
./agy-proxy --port 8000

# Windows:
agy-proxy.exe --port 8000

Option 2: Python / UV

git clone https://github.com/kqlio67/agy-proxy.git
cd agy-proxy
uv run python main.py --port 8000

Open http://localhost:8000 in your browser to access the Web Dashboard!

---

📄 License & Terms of Service

Distributed under the GNU General Public License v3.0 (GPLv3).
This software is developed independently for educational, research, and API interoperability purposes only. Users are responsible for complying with the Terms of Service of the respective platforms.