---
title: RustySqueezer
url: 'https://niklas-stephan.de/en/zoo/rusty-squeezer'
markdown: 'https://niklas-stephan.de/en/zoo/rusty-squeezer.md'
lang: en
date: '2026-10-28'
description: 'When an agent conversation outgrows the context window, the usual fix is to ask an LLM to summarize it — slow, lossy, and unpredictable. RustySqueezer instead compresses deterministically: structure-aware trimming of tool output, old turns and duplicated content, so the conversation keeps its shape…'
---

Rust · axum

# RustySqueezer

Deterministic, model-free context compaction behind an OpenAI-compatible API: squeezes long agent conversations into the context window without asking an LLM to summarize them.

2026-10-28

When an agent conversation outgrows the context window, the usual fix is to ask an LLM to summarize it — slow, lossy, and unpredictable. RustySqueezer instead compresses deterministically: structure-aware trimming of tool output, old turns and duplicated content, so the conversation keeps its shape and the important parts stay verbatim.

It sits behind an OpenAI-compatible API, so any client that speaks that API can use it as a compaction endpoint — it runs in production here as the compaction model for coding agents, and optionally forwards a narrative summary to a real LLM tier when one is configured.

 [ Repository](https://github.com/handtrixx/RustySqueezer) [ Docs](https://handtrixx.github.io/RustySqueezer/)

---

## Navigation

- Parent: [Zoo](https://niklas-stephan.de/en/zoo.md)
- Previous: [Grava the HUD](https://niklas-stephan.de/en/zoo/grava-the-hud.md)
