
Show HN: Quill Desktop; Lightweight AI harness, no electron/WebKit native GUI
Still a work in progress. Let me know if you like it and want to see it finished. Contributions are welcome.
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
The lightweight, native desktop harness for every LLM
Native Desktop UI APIs (No Electron)
Minimal resource footprint, under 70mb of ram
Extremely fast Rust core
Supports every LLM provider, no lock down
Based on VSCode's design
Downloads
Download from Github releases and run. Uses portable executables so no install is required.
To uninstall delete the executable and the config under $HOME/.config/quill.
Note: the MacOS build is unsigned because I am not spending US$99/year for an Apple developer account
recording-2026-10-25.mp4
How does this work?
Quill uses the work done by the Zed editor, utilizing their cross platform gpui-kit.
Simply put, gpui leverages native APIs (Metal, Vulkan and DirectX) to draw the UI. You can think of it as using a game engine to render a UI, except it only renders frames when things are moving, ensuring power usage / battery life is unaffected.
It renders frames at your screen's refresh rate (e.g. 144fps) so the experience is buttery smooth.
All the current desktop harnesses are built on top of Electron (basically a full-fat dedicated clone of Chrome running only one website).
Conventionally, this is done to minimize the friction associated with cross-platform application development and be competitive in feature delivery. Essentially, sacrifice a little user experience in exchange for the ability to move faster than competitors.
In my view, harnesses have reached a level of maturity now where the features we expect from them are well enough defined. In addition, there is no longer the pressure to be a first mover in the Desktop Agent Harness space.
I think we can now afford a high quality tool that respects your computers resources while also supporting all model providers.
Additionally, the technology powering Quill means that adding features is not so hard that it justifies an Electron build.
UI Icons from Codicons by Microsoft (CC BY 4.0).
UI Icons from Font Awesome Free by Fonticons, Inc. (CC BY 4.0).
UI Icons from Bootstrap Icons by The Bootstrap Authors (MIT).
UI framework GPUI from the Zed editor by Zed Industries (Apache 2.0).
Agent harness goose by the Agentic AI Foundation (Apache 2.0).
License
I have gone with an unconventional non-commercial license that has an clause for AI training. I'm not anti-AI at all (obviously) and have no expectations that my license will be respected.
The intention is that - should this code be included in training data - throw a little something my way for the training data.
If you're doing research or working on an open weight model, feel free to include this in your training data.
This is a weekend project of mine. If you genuinely enjoy it and want to see more, please consider starring the repo and supporting my efforts.
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。