TypeNext v0.4.0

TypeNext 0.4.0

Organise your writing into workspaces, folders and tags, keep every version of a note, and edit selected text with AI that only proposes.

Workspaces

• Separate workspaces, like browser profiles, each with its own notes, folders, tags and contexts. Switch at the bottom left, or with ⌘ ⌥ 1 to 9 (Ctrl+Alt+1 to 9). Each workspace remembers its tabs.
• A shared workspace appears inside every other workspace, for a handbook, templates or reference notes.
• Each workspace can sync to its own Git repository, and can wear an icon and a colour.
• Contexts choose the workspaces that can use them. Your notebook opens as one workspace, Personal, exactly as before.

Folders and tags

• Nested folders with colours. Drag notes and folders to move them; removing a folder keeps its notes.
• Tags under each note's title, tag filters in the sidebar and a tag manager. Renaming a tag to an existing name merges the two.
• A folder can be live context for a note: every note in it, including ones you add later.

History

• Optional versions of every note, on this device and without Git (Preferences → History). Versions are kept automatically after a meaningful change, or only when you save one (⌘ ⌥ S).
• Compare any version with your note word by word, bring back single parts, or restore the whole version. Safety copies are kept before AI edits, restores and large deletions.

Writing

• Edit with AI: select text and choose Polish, Fix grammar, Rewrite, Paraphrase, Make shorter, Make longer or your own request (⌘ J). You see every change before it enters the page. Needs an active chat model.
• Find and replace in a note (⌘ F, ⌘ ⌥ F), with match case, whole word and patterns.
• Links show only their text until the cursor enters them. Link cards add Copy, Edit and Unlink.
• Choose your own keys for suggestions, accepting a word and AI actions (Preferences → Shortcuts).
• Set the most words in a suggestion, and start your instructions from twelve templates.
• A small animation plays while a model works: Orb, Spark, Bloom, Firefly and more, or none.
• Custom palette colours open a colour picker.

Fixes

• Selected text is readable again in dark mode, and the format bar matches the theme.
• Editor tooltips no longer show through dialogs.
• The Updates settings have proper spacing.

Update from 0.3.0

0.3.0 updates itself: a notice offers Restart to update when 0.4.0 has downloaded. Packaging is the same as in 0.3.0: unsigned Windows binaries and ad-hoc-signed Mac bundles. Updates are checked against the TypeNext update key before they install.

TypeNext v0.3.0

TypeNext 0.3.0

TypeNext now keeps itself up to date, and the window looks right again on macOS.

Updates inside the app

• TypeNext checks the GitHub releases shortly after launch and every six hours. A new version downloads quietly in the background and is checked against the TypeNext signing key before it can install.
• A small notice shows when the update is ready, with What’s new, Later and Restart to update. Your notes are saved first. Nothing restarts without your choice.
• Preferences → Updates shows your version, Check now, and switches for automatic checks and background downloads. The command palette has Check for updates.
• The notice stays hidden in zen mode and while a dialog is open.

Fixes

• macOS: the sidebar subtitle "a place for your words" no longer overlaps the TypeNext name below the window buttons.
• The colour palette cards in Preferences → Writing have equal sizes and line up again.
• Messages float above the status bar instead of covering it, and close sooner.
• A narrow page keeps the word count and the save state on one line, and a long note title ends with an ellipsis.

Update from 0.2.0

Version 0.2.0 cannot update itself. Download 0.3.0 once from the releases page. From 0.3.0 on, updates install from inside the app.

Packaging is the same as in 0.2.0: unsigned Windows binaries and ad-hoc-signed Mac bundles, so the system may show its normal trust prompt for the first install. In-app updates are signed separately with the TypeNext update key.

TypeNext v0.2.0

TypeNext 0.2.0

You write. AI suggests. This release makes the editor richer, makes suggestions understand your sources by meaning, and adds long dictation on your device. Everything new that touches your notes runs on your computer unless you turn on a service.

Writing

• Notes start at once: Ctrl/⌘ N and type. The first line becomes the title until you type one.
• A real Markdown editor: live bullets, task boxes and tables. Tab moves between table cells and indents list items. Enter continues a list.
• The / insert menu adds headings, lists, tables, code, links, dates and your own snippets. You can change the trigger character or turn the menu off.
• [[Note links]] complete note titles, show a hover card, and open with a Ctrl/⌘ click. A missing note can be created from its link. Web links show a preview card.
• Writing flow: a format bar on selection, smart typography (-- becomes —), typewriter scrolling, paragraph or sentence focus, and a pasted web address becomes a link with the page title.
• A quiet outline in the margin, the notes that link to the current note, reading time and an optional daily word goal.
• The command palette (Ctrl/⌘ P) finds notes, headings in the note and every action.
• Three zen levels: hide the side panels, hide the toolbars, or full screen. The note graph (Ctrl/⌘ Shift G) shows how notes and contexts connect.

Suggestions

• Semantic search with Google EmbeddingGemma 2, beside BM25. In the desktop app, TypeNext downloads a pinned llama.cpp runtime and the model, checks each file and runs it on your computer. It also understands images, audio and video in your references.
• Suggestions follow the structure at the cursor: a heading stays a heading, a list item stays one item, a table cell gets a few words. Every suggestion shows the reference it came from.
• Suggestions wait while you type fast and come sooner at a pause. You can set the delays in Preferences → Suggestions → Timing.
• Cloudflare Workers AI is a new external provider. OpenAI, Anthropic and Cloudflare have a model list to choose from.
• Web facts (opt-in): select text, press Ctrl/⌘ Shift K, and insert a fact with its source from an OpenAI web search.

Dictation

• Fluid Speech Mini by FluidVoice (CC BY 4.0) for long English dictation in the desktop app, through a pinned parakeet.cpp runtime. Finished phrases enter the note at each pause.
• A small chip at the cursor shows the level, the time and the state. It never covers your text or takes the focus. Voice commands add punctuation, lines and list items. Dictation can stop by itself after silence.
• Whistle stays the small default (about 17 MB, 7 languages, clips of up to 30 seconds).

Sync and setup

• Optional Git sync keeps every note as a Markdown file in a repository of your choice, with the Git installed on your computer. When both sides change, both versions are kept.
• If Hugging Face asks for a sign-in for a model download, TypeNext shows a short guide and keeps the read token only for huggingface.co.
• New Preferences and Contexts screens. On macOS, the window uses the native traffic lights.

Packaging is the same as in 0.1.3: unsigned Windows binaries and ad-hoc-signed Mac bundles, so the system may show its normal trust prompt. Semantic search and Fluid Speech Mini need the desktop app and a one-time download. Model output is optional and can be wrong; check facts before you rely on them.

TypeNext v0.1.3

TypeNext 0.1.3

The first public TypeNext release: a local-first Markdown notebook for Windows x64, Apple Silicon Macs and Intel Macs. TypeNext-owned code is released under MIT; included dependencies retain their upstream licenses.

• A quiet, tabbed writing space with neutral dark and light palettes, custom colours, adjustable fonts and a focus mode.
• Local autosave, recent notes, Markdown file import/export, and recovery from a valid saved backup. See the reliability documentation for the remaining unsaved-edit window.
• Reusable named context packages containing documents, PDFs, text, Markdown, websites or live links to other notes. Several packages can inform one note, and individual sources can be excluded.
• A shipped fictional Bramble House sample demonstrates context-grounded offline suggestions. The default local recall/starter engine needs no API key, model download or server.
• Ctrl/Cmd+Space offers suggestions at the cursor. Hosted manual choices provide up to three fresh alternatives from one request; Tab accepts the preview and undo restores the writing.
• Optional downloaded SmolLM2 suggestions, a loopback local model server, or deliberately activated continuous external suggestions through OpenRouter, OpenAI, Anthropic or a custom endpoint. A local failure never switches to a hosted provider.
• Offline Whistle dictation after an explicit model download: Ctrl/Cmd+Shift+D starts/stops recording, microphone selection is available, typing can continue, and the transcript is inserted at its tracked position before suggestions resume. Recording is limited to 30 seconds; failed insertions remain recoverable.

Windows downloads include a setup EXE, MSI and a portable ZIP containing the application EXE. Each Mac architecture has a DMG and a ZIP containing TypeNext.app. Application archives include the project license, third-party notices, font/runtime licenses and applicable native corresponding source. SHA256SUMS.txt covers the binary distribution assets.

This initial workflow produces unsigned Windows binaries and ad-hoc-signed Mac bundles; it does not claim Developer ID notarization or Windows certificate signing. macOS and Windows may show their normal trust prompts. The portable Windows build requires Microsoft WebView2 Runtime. Optional model downloads require a network connection for setup and then run locally. In-app automatic updating is not enabled in this release.

Model output remains optional and may be irrelevant or inaccurate despite supplied context. The real free-provider demo check reached the provider with the correct arrival reference, but its returned suggestion did not retain those facts; this release does not claim verified factual accuracy for hosted suggestions. PDF extraction reads available text rather than OCR of scanned pages.