How it works
Jolly runs locally on your PC. If you feel like a text needs spell checking, just copy it and hit Enter. Jolly reads your clipboard, passes it through a local LLM, and pastes the corrected text back. Your machine is slow? Use API keys or conventional grammar checking.
Why it exists
Spell checkers are annoying — squiggly lines and too much clicking. Nowadays I tend to just paste text into an AI with the prompt "fix spelling." A lot of times I feel uneasy about my mails and notes being collected by LLM providers. So why not do it locally and make it fun?
Under the hood
Built with SvelteKit, Tailwind, Tauri, and llama.cpp — in Rust and TypeScript. You can select and download your favourite model in the app. If your machine doesn't allow local AI inference, you can also use your API keys. Jolly misuses Harper by just picking the most likely fix. Everything runs locally — nothing leaves your machine.
Open source
Free to download, free to use, free to clone, free to improve. Read the source if you're curious.
Made to learn
Jolly started as a way to get to grips with frontend development. Inkscape, SvelteKit and Tailwind made that a brezze. Spell checking is a piece of cake for the big boy LLMs. Running locally in a reasonable inference time on the other hand, is quite a challenge. For one-shot spell checking, without clicking on lints, LLMs show superior results at a high price. On the other hand, conventional grammar checkers are super fast and work just as well (mostly). AI is amazing but is it always the best choice?
The source is open. If something looks wrong or could be better, pull requests are welcome.