dshakes/distil

Overview

A Python library for LLM context compression that prunes context causally and is cache-aware. It provides a quality contract, certified non-inferior across seven domains, and works with any SDK. Designed for agentic runtimes.

Best for

Best for
Developers building agentic systems needing safe context compression

Use cases

• Compressing conversation history for agent loops
• Reducing token usage in multi-step LLM calls
• Maintaining output quality while shrinking context

How to use

Install

uvx --from distil-llm distil bench      # certify savings + quality across 9 domains, in seconds

Tested with

Claude Desktop, Claude Code, Cursor, Windsurf, Cline, Continue, VS Code, ChatGPT

Notes

A Python library for LLM context compression that prunes context causally and is cache-aware. It provides a quality contract, certified non-inferior across seven domains, and works with any SDK. Designed for agentic runtimes.

16 stars on GitHub. Last updated 2026-09-02.

Use cases

• Compressing conversation history for agent loops
• Reducing token usage in multi-step LLM calls
• Maintaining output quality while shrinking context

Pros

• Cache-aware compression reduces redundant processing
• Certified non-inferior quality across seven domains
• SDK-agnostic integration

Cons

• Small community adoption with only 16 stars
• Python only, no other language bindings
• Quality certification limited to seven domains

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.