sievellm 0.2.0

Smart AI router — filters requests to the right model, saving 40-80% on costs

Sieve

Smart AI router — filters requests to the right model.

🚧 Early Development — Works, but evolving. Feedback welcome!

Stop overpaying for AI. Sieve automatically routes simple queries to cheap models and complex ones to capable models, saving you 40-80% on API costs.

Installation

pip install sievellm

Quick Start

from sieve import route

# Simple query → routes to cheap model
response = route("What is 2+2?")
print(response)  # 4
print(f"Cost: ${response.cost_usd:.6f}")  # ~$0.000001

# Complex query → routes to capable model  
response = route("Analyze the trade-offs between microservices and monoliths")
print(f"Cost: ${response.cost_usd:.4f}")  # ~$0.02

Setup

Set your OpenAI API key:

export OPENAI_API_KEY="sk-..."

How It Works

1. Analyzes your prompt for complexity signals
2. Routes to the cheapest model that can handle it
3. Tracks actual cost per request

Try It

git clone https://github.com/krymsnax/sieve
cd sieve
pip install -e .
python3 chat.py

Roadmap

• Anthropic Claude support
• Streaming responses
• Budget limits
• ML-based complexity detection

License

MIT