Antigravity Server

Production-grade AI Gateway — OpenAI-compatible API for Gemini & Claude

English · Русский · Upstream 中文

---

Headless Rust server that transforms Google AI Studio and Anthropic Console web sessions into standard OpenAI-compatible APIs. Deploy on VPS, run as systemd daemon, manage via CLI or Web UI.

Quick Start

# Build
cargo build --release -p antigravity-server

# Run
./target/release/antigravity-server
# → http://127.0.0.1:8045

import openai

client = openai.OpenAI(
    api_key="sk-antigravity",
    base_url="http://127.0.0.1:8045/v1"
)

response = client.chat.completions.create(
    model="gemini-3-pro-high",
    messages=[{"role": "user", "content": "Hello!"}]
)

# Claude Code CLI
export ANTHROPIC_API_KEY="sk-antigravity"
export ANTHROPIC_BASE_URL="http://127.0.0.1:8045"
claude

---

Why This Fork?

Capability	Upstream (Tauri)	This Fork (Axum)
Deployment	Desktop GUI	Headless VPS daemon
Frontend	React + TypeScript	Leptos (Rust → WASM)
Rate Limiting	Reactive (retry on 429)	AIMD predictive
Account Rotation	Standard retry	Smart exclusion-based rotation
Reliability	Basic failover	Circuit breakers + health scores
Persistence	Direct file I/O	Actor loop (race-free)
Observability	Local UI only	Prometheus metrics + REST API
Audio/Video	Not supported	Full multimodal

---

Smart Account Rotation

When running multiple accounts (e.g., 15 Google AI Studio accounts), efficient load distribution becomes critical under high concurrency. This fork implements exclusion-based rotation — each request maintains memory of which accounts already failed, ensuring retries always try fresh accounts.

Request lifecycle:
┌─────────────────────────────────────────────────────────────┐
│ Attempt 1: Account A (ultra-tier) → 503 Service Unavailable │
│            attempted = {A}                                   │
├─────────────────────────────────────────────────────────────┤
│ Attempt 2: Account B (pro-tier) → 429 Rate Limited          │
│            attempted = {A, B}                                │
├─────────────────────────────────────────────────────────────┤
│ Attempt 3: Account C (free-tier) → 200 OK ✓                 │
│            Response returned to client                       │
└─────────────────────────────────────────────────────────────┘

Benchmark (50 concurrent requests):

Metric	Result
Accounts utilized	14 different accounts
Rotation events	362 distributed across pool
Success rate	100% (50/50 HTTP 200)
Total time	18.9 seconds

The exclusion set is per-request (no cross-request contamination), thread-safe (stack-allocated HashSet), and has minimal overhead (~6KB worst case for 14 accounts).

---

Architecture

┌─────────────────┐     ┌─────────────────────┐     ┌──────────────────┐
│  Claude Code    │     │                     │     │  Google Gemini   │
│  OpenAI SDK     │ ──▶ │  Antigravity Proxy  │ ──▶ │  Anthropic API   │
│  Cursor / IDE   │     │   (localhost:8045)  │     │  (via OAuth)     │
└─────────────────┘     └─────────────────────┘     └──────────────────┘

Endpoints:

• POST /v1/chat/completions — OpenAI-compatible chat
• POST /v1/messages — Anthropic-compatible messages
• GET /v1/models — Available models
• POST /v1/images/generations — Imagen 3 (DALL-E compatible)
• POST /v1/audio/transcriptions — Whisper-compatible

Resilience API:

• GET /api/resilience/health — Account availability
• GET /api/resilience/circuits — Circuit breaker states
• GET /api/resilience/aimd — Rate limit telemetry
• GET /api/metrics — Prometheus metrics

---

Multimodal

Audio (official OpenAI format):

response = client.chat.completions.create(
    model="gemini-3-pro",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "Transcribe this audio"},
            {"type": "input_audio", "input_audio": {"data": audio_b64, "format": "wav"}}
        ]
    }]
)

Video (Gemini extension):

response = client.chat.completions.create(
    model="gemini-3-pro",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "Describe this video"},
            {"type": "video_url", "video_url": {"url": f"data:video/mp4;base64,{video_b64}"}}
        ]
    }]
)

Supported: wav, mp3, ogg, flac, m4a | mp4, mov, webm, avi

---

CLI

antigravity-server account list          # List accounts with quotas
antigravity-server account add --file x  # Add account from JSON
antigravity-server account refresh all   # Refresh all quotas
antigravity-server warmup --all          # Warmup sessions
antigravity-server status                # Proxy statistics
antigravity-server config show           # Current config

---

Deployment

Systemd (user service):

# ~/.config/systemd/user/antigravity.service
[Unit]
Description=Antigravity AI Gateway
After=network.target

[Service]
ExecStart=%h/.cargo/bin/antigravity-server
Restart=always
Environment=RUST_LOG=info
Environment=ANTIGRAVITY_PORT=8045

[Install]
WantedBy=default.target

systemctl --user enable --now antigravity

Environment:

Variable	Default	Description
ANTIGRAVITY_PORT	8045	Server port
ANTIGRAVITY_DATA_DIR	~/.antigravity_tools	Data directory
RUST_LOG	info	Log level

---

Project Structure

crates/
├── antigravity-types/    # Shared types, errors, models
├── antigravity-core/     # Business logic (proxy, AIMD, circuits)
└── antigravity-client/   # Rust SDK (auto-discovery, streaming)

antigravity-server/       # Axum HTTP server + CLI
src-leptos/               # Leptos WASM frontend
vendor/antigravity-upstream/  # Upstream reference (submodule)

---

License

Based on lbjlaq/Antigravity-Manager.

License: CC BY-NC-SA 4.0 — Non-commercial use only.

Built with Rust