lessllm 0.2.0

A lightweight LLM API proxy framework with network connectivity and usage analysis

LessLLM v0.2.0

🚀 A lightweight, enterprise-grade LLM API proxy framework with comprehensive analytics and intelligent routing

Do more with less code/gpu/mem

✨ Key Features

🎯 Intelligent API Routing

• Smart Format Conversion: Seamlessly switch between Claude Messages API and OpenAI Chat Completions API
• Provider Transparency: Use any API format with any provider (Claude ↔ OpenAI)
• Automatic Routing: Based on model names and endpoint types
• Streaming Support: Real-time format conversion for streaming requests

📊 Enterprise Analytics

• 100% Data Capture: Complete HTTP request/response logging to DuckDB
• Performance Monitoring: TTFT, TPOT, throughput analysis
• Cost Tracking: Token usage and cost estimation per request
• Cache Analytics: Cache hit rates and savings calculation
• Interactive Dashboard: Web-based analytics with real-time data

🔧 Multi-Provider Support

• Claude (Anthropic): Direct Messages API support + Aliyun proxy compatibility
• OpenAI: Chat Completions API with full feature parity
• Extensible: Easy to add new providers
• Load Balancing: Intelligent provider selection

🌐 Web Interface

• Real-time Monitoring: Live request tracking and analysis
• Interactive Tables: Click-to-view detailed request information
• Data Visualization: Charts and graphs for usage patterns
• SQL Query Interface: Custom analytics with pre-built templates

Quick Start

Installation

pip install -e .

Initialize Configuration

lessllm init --output lessllm.yaml

Start the Proxy Server

# Set your API keys
export OPENAI_API_KEY="your-openai-key"
export ANTHROPIC_API_KEY="your-claude-key"

# Start server
lessllm server --config lessllm.yaml --port 8000

Start the Analytics Dashboard GUI

# Start GUI (requires optional GUI dependencies)
pip install -e .[gui]
lessllm gui --port 8501 --host localhost

Use with OpenAI Client

import openai

# Point OpenAI client to LessLLM proxy
client = openai.OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="dummy"  # LessLLM uses configured keys
)

response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Hello!"}]
)

Configuration

Create a lessllm.yaml configuration file:

proxy:
  socks_proxy: "socks5://127.0.0.1:1080"  # Optional
  timeout: 30

providers:
  openai:
    api_key: "${OPENAI_API_KEY}"
  claude:
    api_key: "${ANTHROPIC_API_KEY}"

logging:
  enabled: true
  storage:
    type: "duckdb"
    db_path: "./lessllm_logs.db"

analysis:
  enable_cache_estimation: true
  enable_performance_tracking: true

Project Status

Current Implementation Status:

✅ Core Framework

• ✅ Configuration management system
• ✅ FastAPI proxy server
• ✅ Network proxy support (HTTP/SOCKS)
• ✅ Provider abstraction layer

✅ API Support

• ✅ OpenAI provider implementation
• ✅ Claude provider implementation
• ✅ Streaming and non-streaming support

✅ Logging & Analysis

• ✅ DuckDB storage system
• ✅ Performance tracking (TTFT/TPOT)
• ✅ Cache estimation algorithms
• ✅ Cost calculation utilities

✅ Developer Tools

• ✅ CLI interface
• ✅ Example configurations
• ✅ Test client scripts
• ✅ Analytics dashboard GUI

Next Steps

🔄 In Progress:

• Unit tests and integration tests
• Documentation improvements
• Performance optimizations

📋 Planned:

• More LLM provider integrations
• Advanced caching strategies
• Batch request optimization

License

MIT License