🚀 Tokenxygen

Universal token optimizer for coding agents — save 40-70% on LLM costs.

Tokenxygen is a drop-in proxy that sits between your coding agent (Claude Code, Cursor, Windsurf, Codex) and LLM APIs, automatically optimizing tokens in-flight. Zero code changes required.

┌─────────────┐     ┌────────────────┐     ┌──────────┐
│  Coding      │────▶│  Tokenxygen    │────▶│  OpenAI  │
│  Agent       │◀────│  Proxy         │◀────│  API     │
└─────────────┘     └────────────────┘     └──────────┘
                    🎯 Cache | Compress | Route | Budget

⚡ Quick Start

pip install tokenxygen

# Start the proxy
tokenxygen serve

# In another terminal — use your coding agent normally
export OPENAI_BASE_URL=http://127.0.0.1:8420/v1
claude  # or cursor, windsurf, etc.

That's it. Tokens are optimized automatically.

🔧 How It Works

Tokenxygen applies multiple optimization strategies in order:

#	Strategy	What It Does	Typical Savings
1	Semantic Cache	Reuse identical/similar responses	30-60%
2	File Deduplication	Remove repeated file contents	10-30%
3	Plugin Strategies	Custom compression (diffs, imports, tool outputs)	5-20%
4	LLMLingua-2	Token-level compression (when installed)	40-60%
5	Smart Routing	Send simple queries to cheap models	40-70%
6	Budget Guards	Enforce daily limits, auto-downgrade	Variable
7	Failover	Automatic provider switching on failure	Availability

Combined, these typically save 40-70% on token costs without noticeable quality loss.

📊 Dashboard

Start the proxy with a real-time web dashboard:

tokenxygen serve --dashboard-port 8421
# Open http://127.0.0.1:8421

🐳 Docker Deployment

# Quick start with Docker Compose
cp .env.example .env
# Edit .env with your API keys
docker compose up -d

# Services:
# - Proxy: http://localhost:8420
# - Dashboard: http://localhost:8421
# - Prometheus: http://localhost:9091

🛡️ Security Features

Tokenxygen includes comprehensive security hardening:

Feature	Description
SSRF Prevention	Blocks requests to localhost, private IPs, and non-HTTP protocols
Rate Limiting	Per-IP sliding window with Redis support for distributed deployments
Security Headers	CSP, HSTS, X-Content-Type-Options, X-Frame-Options, and more
Admin Authentication	API key protection for sensitive endpoints
Request Validation	Path traversal and null byte injection prevention
Size Limits	10MB maximum request body
Secure Logging	Authorization headers never logged; correlation IDs for audit trails

Environment Variables

# Security settings
export TOKENXYGEN_ADMIN_KEY="your-admin-api-key"  # Protect admin endpoints
export TOKENXYGEN_REDIS_URL="redis://localhost:6379/0"  # Distributed rate limiting

# Production settings
export OPENAI_BASE_URL="http://localhost:8420/v1"
export OPENAI_API_KEY="sk-..."

📦 Installation

# Basic install
pip install tokenxygen

# With LLMLingua-2 for deeper compression
pip install tokenxygen[llmlingua]

# Full install (all features)
pip install tokenxygen[full]

# Development
pip install -e ".[dev]"

🛠️ CLI Commands

# Start the proxy server + dashboard
tokenxygen serve
tokenxygen serve --json-logs --log-level DEBUG

# View token usage stats
tokenxygen stats

# Show daily cost history
tokenxygen history --days 30

# Count tokens in text
tokenxygen tokens "Hello, world!"

# Show compression results
tokenxygen compress "your code here"
tokenxygen deep-compress "long text here"

# List available models and costs
tokenxygen providers

# Compare cost between two models
tokenxygen compare gpt-4o gpt-4o-mini --tokens 5000

# Show cheapest model for different sizes
tokenxygen cheapest

# Set budget limit
tokenxygen budget 20.00

# Check budget status
tokenxygen budget-status

# Cache statistics
tokenxygen cache-stats

# List loaded plugins
tokenxygen plugins

# Show failover pool status
tokenxygen failover

# Show metrics
tokenxygen metrics

# Get setup instructions
tokenxygen setup

# Print shell exports
eval "$(tokenxygen env)"

# Reset analytics
tokenxygen reset --clear-cache

🔌 Multi-Provider Support

Provider	Models	Cost
OpenAI	GPT-4o, GPT-4o-mini, GPT-4-turbo	$0.15-30/1K tokens
Anthropic	Claude 3.5 Sonnet, Claude 3 Haiku	$0.25-75/1K tokens
Google	Gemini 1.5 Pro, Gemini 1.5 Flash	$0.075-10.5/1K tokens
Ollama	Llama 3.1, CodeLlama, DeepSeek Coder	Free (local)

🛡️ Budget Guards

Utilization	Action	Effect
< 80%	PASS	No changes
80-95%	DOWNGRADE	Route to cheaper model
95-100%	AGGRESSIVE_COMPRESS	Maximum compression
> 100%	BLOCK	Return 429 error

📊 Monitoring

Prometheus Metrics

# Metrics endpoint
curl http://localhost:8420/metrics

# JSON metrics
curl http://localhost:8420/metrics/json

Response Headers

X-Tokenxygen-Original-Tokens: 12450
X-Tokenxygen-Optimized-Tokens: 4320
X-Tokenxygen-Saved: 8130
X-Tokenxygen-Strategies: file_dedup,plugin:compress_diffs

⚙️ Configuration

from tokenxygen.config import settings

# Core settings
settings.cache.enabled = True
settings.compress.aggressiveness = 0.4
settings.router.enabled = True
settings.budget.daily_limit_usd = 20.0

# Security settings
settings.proxy.enable_hsts = True  # Enable when behind HTTPS proxy
settings.proxy.cors_origins = ["https://yourdomain.com"]  # CORS origins
settings.proxy.trust_forwarded_headers = True  # When behind reverse proxy
settings.proxy.redis_url = "redis://localhost:6379/0"  # Distributed rate limiting

🧪 Development

git clone https://github.com/Dream-Pixels-Forge/tokenxygen.git
cd tokenxygen
pip install -e ".[dev]"

# Run tests
pytest

# Run with auto-reload
tokenxygen serve --reload

🗺️ Roadmap

• v0.1 — Core proxy, cache, compression, analytics
• v0.2 — Smart routing, budget guards, web dashboard
• v0.3 — Multi-provider support, plugin system
• v0.4 — LLMLingua-2 compression, failover, load balancing
• v0.5 — Production hardening, metrics, Docker, security
• v1.0 — Production ready (rate limiting, auth, API keys)

📄 License

MIT