Token Optimization

Save 30-50% on OpenAI & Anthropic API Costs

Drop-in replacement clients with automatic caching, compression & GPU acceleration.

---

Install

git clone https://github.com/DeadManOfficial/token-optimization.git
cd token-optimization
pip install -r requirements.txt

# GPU support (optional)
pip install torch --index-url https://download.pytorch.org/whl/cu118

---

Quick Start

from src.auto_optimizer import OptimizedAnthropic, OptimizedOpenAI

# Anthropic (Claude)
client = OptimizedAnthropic(use_gpu_embeddings=True)
response = client.messages.create(
    model="claude-sonnet-4-5-20250929",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Hello!"}]
)

# OpenAI (GPT)
client = OptimizedOpenAI(use_gpu_embeddings=True)
response = client.chat.completions.create(
    model="gpt-4-turbo",
    messages=[{"role": "user", "content": "Hello!"}]
)

# Check savings
print(client.get_stats())

---

Features

Feature	Description
Drop-in Replacement	No code changes needed
Smart Caching	MD5 + semantic deduplication
GPU Acceleration	CUDA-powered semantic search
Multi-Provider	Anthropic & OpenAI support
Analytics	Track savings & performance

---

Performance

Configuration	Savings	Use Case
CPU Caching	20-30%	General usage
GPU Semantic	40-50%	Agent frameworks
Mixed Workload	30% avg	Production apps

---

GPU Requirements

• NVIDIA GPU (RTX recommended)
• CUDA 11.8+
• 4GB+ VRAM

Falls back to CPU automatically if unavailable.

---

Architecture

src/
├── auto_optimizer.py       # Drop-in API wrappers
├── token_optimizer.py      # 7 optimization techniques
└── gpu_embeddings.py       # GPU semantic caching

---

Related

• mcp-auditor — Security & compliance auditor for Claude
• claude-canvas — External monitor for Claude Code
• AI-Updates — Daily AI intelligence briefs

---

License

MIT

---

_{BUILD > BUY}