How I achieved 3.7x less memory usage than Cursor by ripping out Electron
这条记录涉及编程工具或代码能力更新，适合开发者评估工作流变化和可复用价值。
一位开发者基于其在高性能系统架构和底层优化方面的经验，针对现代AI编辑器内存占用过高的问题，开发了一款名为Axiom的编辑器。Axiom通过移除VSCode OSS中的Electron框架，转而使用LaVista运行编辑器，从而显著降低了内存消耗。与Cursor相比，Axiom的内存使用量减少了3.7倍，比VSCode减少了33%。…
https://preview.redd.it/m9dr0yr2snbh1.png?width=1080&format=png&auto=webp&s=83d9ed88235f8ec7186ee5df1885768ed73cab60
My background is in high-performance systems architecture and low-level optimization, and recently, the memory bloat in modern AI editors has been driving me crazy (WE OBVIOUSLY CAN DO BETTER, WHY ARENT WE???)
So, I decided to build something significantly leaner and minimal. I built Axiom which uses up to 3.7x less memory than Cursor and 33% less than VSCode.
To hit this benchmark, I took VSCode OSS and stripped Electron out completely. Instead of relying on the bundled Chromium instance, I made the editor run inside LaVista ( https://github.com/IASoft-PVT-LTD/LaVista ).
This allowed me to drop the footprint of three idle windows down to just 759 MB, compared to the 2,802 MB you'd see in Cursor.
- AxiomAI: A Bring Your Own Key (BYOK) setup with a local autocomplete and local router system.
- Token Management: Built-in tracking to monitor, analyze, and set hard limits on your API token usage so you never get a surprise bill.
- FlowViz: A native visualization engine that lets you render plots, flowcharts, and fully interactive 3D scenes directly in the editor.
I am currently rolling out the beta and would love for some technical folks to try it out and try to break it.
You can check it out and register for the beta here: https://iasoft.dev/software-engineering/products/axiom/
Would def love to hear your thoughts on the native webview approach or answer any questions about the LaVista implementation!