Hi Whale bro! ^_^ @Hmbown,
I came across your work and thought you might be interested in a small library I've been working on: mosaic-compress.
It's a stateless dialogue compression tool that mimics human memory — it keeps LLM conversations bounded forever without session management, so theoretically it can support unlimited conversation rounds. I think this could be especially useful for long coding sessions where context keeps growing.
The compression is configurable via parameters like lightStart, lightWindow, heavyStart, and heavyWindow, so you can tune it to your needs. It's also LLM-agnostic — you bring your own callLLM function (OpenAI, Anthropic, local models, etc.).
The repo is under the MIT License, so feel free to use it however you like — no strings attached.
Git address: https://github.com/TuringCorp-net/mosaic_compress
If you find it interesting, a star would be much appreciated 😊
Thanks for your time!
Hi Whale bro! ^_^ @Hmbown,
I came across your work and thought you might be interested in a small library I've been working on: mosaic-compress.
It's a stateless dialogue compression tool that mimics human memory — it keeps LLM conversations bounded forever without session management, so theoretically it can support unlimited conversation rounds. I think this could be especially useful for long coding sessions where context keeps growing.
The compression is configurable via parameters like lightStart, lightWindow, heavyStart, and heavyWindow, so you can tune it to your needs. It's also LLM-agnostic — you bring your own callLLM function (OpenAI, Anthropic, local models, etc.).
The repo is under the MIT License, so feel free to use it however you like — no strings attached.
Git address: https://github.com/TuringCorp-net/mosaic_compress
If you find it interesting, a star would be much appreciated 😊
Thanks for your time!