Offline English grammar and spell checker that runs locally in browsers, editors and a desktop app.
Harper checks English grammar in milliseconds on your own device, with no cloud and no LLM.
Harper is an English grammar checker written in Rust that runs entirely on your device. It suits writers, students and developers who want Grammarly-style checks without sending text to a server.
Harper lints English text for spelling, capitalization, awkward phrasing and grammar, and shows suggestions as you type. It ships as a language server, a JavaScript library (harper.js), a Rust crate, browser extensions, editor plugins and a desktop app.
It suits people who care about privacy and speed, including developers who want grammar checks in comments and Markdown in their editor. It does not suit anyone who needs languages other than English or AI rewriting of whole paragraphs.
Every check happens on your device with no cloud round-trip, no telemetry and no LLM in the loop, per writewithharper.com.
harper-ls plugs into VS Code, Neovim, Helix, Emacs and Zed to lint prose and code comments.
Chrome, Edge and Firefox extensions, an Obsidian plugin, a WordPress block-editor plugin and a macOS desktop app.
Small enough to load via WebAssembly, so the checker runs in a web page.
A JavaScript library and Rust crate let you embed the grammar engine in your own app.
The README claims lint times in milliseconds and under 1/50th of LanguageTool's memory footprint.
Harper is free under the Apache-2.0 license, and writewithharper.com shows no paid plan. The only cost is your own device.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
These reviewed AI tools do the same job instantly — from our scored directory.
According to writewithharper.com, every check happens locally with no cloud round-trips and no telemetry.
English only. The README says the core is extensible but other languages need contributions.
The README says LanguageTool needs gigabytes of RAM and a ~16GB n-gram dataset, while Harper uses far less memory and lints in milliseconds.
See Zekai first in Google