NeverWrite

What it is

NeverWrite is a local‑first, multi‑pane desktop application that lets you write Markdown‑style documents while chatting with a variety of AI agents (Claude, Grok‑‑‑…). It is designed as a single place for a personal knowledge‑base (a vault) and the AI‑driven workflows that surround it – writing, research‑oriented navigation, code‑generation / review …​

Why it matters

• Privacy‑first – All data stays on your machine; the app never sends your vault contents to a remote server.
• Unified workflow – You can edit many file types (Markdown, Mermaid‑diagrams etc.) and run an AI chat side‑by‑side, then review the changes the model suggests before they are applied.
• Agent‑agnostic – Works with a growing list of LLM‑back‑ends (OpenAI‑compatible, Anthropic, Mistral, Pi, etc.) and lets you switch providers in the settings.
• Knowledge graph navigation – Wikilinks, backlinks, tags and 2‑/3‑D graph views let you explore the connections inside your vault, a feature that is useful for research or long‑form writing.
• Web clipping – A companion browser extension can capture pages or selections directly into your vault, keeping the whole research pipeline local.

Who it is for

• Writers, researchers, or developers who keep a personal knowledge base (think Obsidian‑style vaults) and want to augment their work with AI assistance without giving up control of their data.
• Teams that prefer a self‑hosted or offline‑first solution for AI‑assisted documentation, code review, or brainstorming.
• Anyone who likes to experiment with multiple LLM providers and wants a single UI to compare them.

How to get started

1. Clone the repo and install the desktop app’s dependencies:

   git clone https://github.com/jsgrrchg/NeverWrite.git
   cd NeverWrite/apps/desktop
   npm install
   npm run dev   # launches the dev version
   

2. Set up an AI runtime – In the app’s settings choose a provider (Claude, Grok, etc.). Some providers need a CLI login or an API key; the README points to a separate AI Runtime Setup doc for Pi.
3. Create or open a vault – A vault is just a folder on your disk that holds Markdown, images, PDFs, Excalidraw files, etc.
4. Start a session – Open an agent chat, attach files or snippets, and let the model suggest edits.
5. Review changes – Use the built‑in review pane to accept or reject whole files or individual hunks before they are written to disk.

Technical highlights

• Frontend – Built with modern web tech (TypeScript, React) and packaged as a native desktop app via Electron (or similar).
• Backend – A Rust‑based native layer handles file I/O, graph indexing, and the bridge to the AI runtimes.
• Cross‑platform – Runs on macOS 13+, Windows, and Linux (both x64 and ARM64).
• Extensible agents – Supports any model that can be accessed through a CLI or an OpenAI‑compatible API; the Pi integration ships with an ACP adapter.
• Testing – Comes with npm‑based unit tests for the desktop app, pnpm checks for the web‑clipper, and cargo tests for the Rust code.

License

Apache‑2.0 – you can use, modify, and redistribute the software under the terms of that permissive license.

---

All information above is taken directly from the repository’s README.