session-compression

What problem does it solve?

AI session compression reduces long multi-turn conversations into compact representations while preserving essential context, enabling longer interactions without losing critical information.

Core Features & Use Cases

• 80-90% token cost reduction through hierarchical and iterative summarization
• 3-20x compression ratios enabling conversations to scale beyond typical context windows
• Multiple strategies (extractive, abstractive, hierarchical, RAG) to balance fidelity and size
• Progressive compression thresholds (70%, 85%, 95%) to adapt to usage and cost
• Anthropic prompt caching integration offering up to ~90% cost savings on cached tokens
• LangChain memory types coverage for seamless integration in Python apps
• Production patterns including checkpointing and resume workflows for durable memory

Quick Start

Install required clients (Anthropic, OpenAI) and instantiate a hybrid memory system combining rolling summarization and retrieval. Feed a multi-turn conversation and retrieve a compressed context suitable for a new user query. Example: initialize memory, add turns, request context.