A quiet, local-first Markdown notepad. A suggestion from your own files waits at the cursor, and Tab keeps it. Or dictate: speech becomes text on your device. Your notes never leave your computer unless you decide.
⬇ Download · Website · User guide · What's new
▶ Watch the 45-second film with sound · recorded in the real app
| ✍️ Your voice first | Suggestions are short and optional. Nothing enters the page until you press Tab. |
| 🔒 Local by default | Recall, the built-in model, semantic search and dictation run on your computer. |
| 🧠 It knows your material | Attach files, PDFs, websites, notes, even images, audio and video. Suggestions use what matters at the cursor. |
| ⚡ Fast and calm | In our tests, each key fits in one frame, even with 400 notes. No toolbars in your way, three levels of zen. |
| 🔗 Notes that connect | [[links]] with hover cards, backlinks, an outline, and a living graph. |
| 🧩 Your choice of model | On-device, LM Studio/Jan/llama.cpp, or OpenRouter, OpenAI, Anthropic and Cloudflare when you turn one on. |
 Type / for headings, lists, tables, code, links, dates and your own snippets. |
 Type [[ to link a note. Hover to preview. ⌘/Ctrl-click to open. |
 Drag a context onto your note. Set the intent once; suggestions follow it. |
 See how your ideas connect with ⌘/Ctrl Shift G. |
 Semantic search on your device with Google EmbeddingGemma 2, beside BM25. |
 Light, dark, contrast and your own palette. |
 ⌘/Ctrl P finds notes, headings and every action. |
 Dictation on your device with Whistle or Fluid Speech Mini. |
Writing
- Start at once. ⌘/Ctrl N and type. The first line becomes the title.
- A real Markdown editor. Live bullets, task boxes and tables. Tab moves between cells and indents lists. Enter continues lists.
- Command palette. ⌘/Ctrl P finds notes, headings and every action.
- Writing flow. A format bar on selection, typewriter scrolling, paragraph or sentence focus, smart typography (
--→ —), a daily word goal, and links from pasted addresses with the page title. - Zen mode. Hide the side panels, hide all toolbars (they return at the edges), or go full screen.
Suggestions that understand your work
- Recall and starters work offline with no setup.
- Hybrid retrieval: BM25 finds matching words; optional EmbeddingGemma 2 finds matching meaning, also in images, audio and video. TypeNext downloads, verifies and runs it locally.
- Structure-aware: a heading stays a heading, a list item stays one item, a table cell gets a few words.
- Your rhythm: TypeNext waits while you type fast and offers sooner at a pause.
- Web facts (opt-in): select text, press ⌘/Ctrl Shift K, and insert a cited fact.
More
- Dictation on your device: Whistle (17 MB), or Fluid Speech Mini by FluidVoice (347 MB) with live phrases, voice commands and auto-stop.
- Git sync: every note as a Markdown file in your repository. Both versions survive a conflict.
- Updates in the app: signed updates from GitHub releases download quietly; TypeNext asks before it restarts.
- Plain files: open and save ordinary Markdown. Autosave keeps a recoverable copy.
Try it in a minute: select Try a sample on the welcome screen, put the cursor after "Guests arrive at Bramble House", press ⌘/Ctrl Space, then Tab.
| Stays on your computer | Leaves only when you turn it on |
|---|---|
| Notes, references, vectors and recordings | An external model: nearby writing and relevant context |
| Recall, the built-in model and local servers | Web facts: the selected text, to OpenAI |
| Semantic search and dictation | Git sync: notes, to your remote |
| API keys (in memory, or the OS vault if you choose) | Link previews: a request to the linked site, no note text |
| Updates: a version check with GitHub, no notes (Preferences → Updates) |
Models download only when you ask, from pinned versions with SHA-256 checks. If Hugging Face asks for a sign-in, TypeNext shows a three-step guide. Details: AI design · built-in models · reliability.
| macOS | Windows / Linux | |
|---|---|---|
| Accept a suggestion · one word | Tab · ⌘ → | Tab · Ctrl → |
| Choices at the cursor | ⌘ Space | Ctrl Space |
| Command palette | ⌘ P | Ctrl P |
| Insert menu · note link | / · [[ |
/ · [[ |
| Web facts | ⌘ Shift K | Ctrl Shift K |
| Dictation | ⌘ Shift D | Ctrl Shift D |
| Zen mode · note graph | ⌘ Shift F · ⌘ Shift G | Ctrl Shift F · Ctrl Shift G |
All shortcuts are in the user guide.
Build from source
Node.js 22.13 or later, stable Rust, and the Tauri prerequisites. Linux also needs libdbus-1-dev.
npm ci
npm run tauri:dev # desktop app
npm run dev # browser preview
npm run check:all # types, lint, unit tests, cargo fmt and clippy
npm run test:e2e # PlaywrightLive provider, embedding and speech tests are opt-in; keys come from a local, untracked .env. See testing, the architecture guide, the editor, semantic search and Git sync.
The website is a Vite and React app in site (npm run site:dev, npm run site:build). The Pages workflow publishes site-dist. The film is made from code; see the film.
Keep the writing flow small, fast and keyboard-friendly. A change to suggestions must test cursor moves, cancellation, the text after the cursor, and the local/external privacy boundary. A change to storage must keep a recoverable copy. Start with CONTRIBUTING.
If TypeNext helps you write, a ⭐ helps other writers find it.
MIT for TypeNext. Built with Tauri, React, CodeMirror and a small BM25 index. Optional models and runtimes are separate downloads with their own licences: llama.cpp (MIT), Google EmbeddingGemma 2 (Apache 2.0), parakeet.cpp (MIT), Fluid Speech Mini by FluidVoice (CC BY 4.0), SmolLM2 and Whistle (Apache 2.0). The website uses React Bits. See the licensing record and third-party notices.