No Type: desktop dictation for macOS and Windows with writing modes
Free desktop dictation for macOS and Windows. Speak messages, emails and notes into your apps with local or cloud speech recognition.
At a glance
- What is it?
- No Type is a desktop dictation app that puts dictated text into whatever field has focus, on macOS and Windows, using either a speech model downloaded to your own machine or a cloud provider with your own API key. Around that core it adds writing modes that reshape the transcript, a personal dictionary, snippets, and a recorder you can pet. The source and documentation were opened under Apache-2.0 in version 0.1.12.
- Who is it for?
- No Type suits somebody who writes long messages, email and notes by voice and wants the transcript cleaned up before it lands in the field, and it suits people who need the recognition to run on their own machine rather than on someone else's servers.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 10 days ago.
- What is it written in?
- Mainly HTML, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on October 22, 2026, and from our analysis. They are not legal advice.
Editorial analysis
Speech into the focused field, and nothing else to learn
No Type turns speech into text for your apps. The mechanism is short: press a shortcut, say what you have in mind, and get text in the focused field.
That framing is the product. There is no document model, no page to open, no document to save. The dictation goes wherever your cursor already is, which means the app is not a writing tool with a speech feature attached, it is a speech front end for whatever you were already writing in. The repository topics match that positioning: desktop-app, dictation, macos, offline, productivity, speech-to-text, transcription, voice-typing, windows and writing-assistant.
Around that core there are three groups of features. Recognition itself, with local speech models downloaded to your computer or a cloud provider connected with your own API key. Shaping, through writing modes that set instructions for messages, emails, notes and tasks, with enhancement, shortening and translation. And the surrounding machinery: empowers to append your own instruction to a single dictation, a personal dictionary, snippets, a history with audio, and insights into how you use it.
The pricing statement is unusually clear for a free application. It is free for personal and professional use, supporting the author is optional, every feature is available without a donation, and cloud providers charge separately under their own pricing. That last clause is the one that matters when you read the feature list, because the local models cost nothing and the cloud path does not.
The project site is notype.tech, the default branch is main, and the primary language of the repository is HTML, which is a fair signal about what this codebase is: a desktop application with a web front end.
The shortcut, the hold-to-talk behaviour and the clipboard fallback
How the recording is triggered is configurable, and the defaults differ between the two platforms in a way worth knowing before you install it.
You hold the shortcut while speaking, or press it to start and stop, and both the shortcut and the recording behaviour can be changed in settings. On macOS the default is the Option key, and on Windows it is a shortcut such as Ctrl + Alt + D, with Right Ctrl as the Windows default: hold it while speaking and release to finish. Updates keep your saved shortcut, so a custom binding survives an upgrade.
The macOS detail that decides whether the app is usable for you is the permission model. Text insertion needs Accessibility permission, and without it the result is copied to the clipboard instead. That is a reasonable fallback rather than a dead end, since you can paste, but it changes the flow from one action to two, and it is the difference between the app feeling instant and feeling like a copy and paste tool.
The Windows behaviour of holding Right Ctrl and releasing to finish is a different design from a toggle, and it suits people who speak in bursts, since the recording ends when the key comes up rather than needing a second press.
Both platforms share the same idea of a small floating recorder that shows what is happening while you talk, which is the feedback loop that makes dictation usable without looking at the app.
Local models or a cloud provider, with your own key
The recognition engine is the choice the user makes at install time, and the README presents it as a table entry rather than a preference.
For local and cloud models, you can download a speech model to your computer or connect a cloud provider using your own API key. The key phrase is your own: the application does not resell access to a speech service, it is a client for a provider account you hold.
The same split appears one layer up. In writing modes, you choose a local or cloud text model to process the transcript after recognition. So a full local setup is possible, with a speech model on disk and a text model on disk, and the README lists offline among the project topics.
That said, the offline label deserves a qualifier. Offline is a property of a configuration, not of the application. Once you point a writing mode at a cloud text model, the transcript leaves the machine on every dictation, and that transcript is whatever you just said, which may include anything you were reading aloud from a screen. The application makes that choice explicit and visible, but the responsibility for it is the user's, and a reader deciding on the app should know which of the two configurations a given mode uses before they dictate something sensitive.
The cost model follows from the same split. The application is free either way. A cloud provider is not, and it bills under its own pricing, so the recurring cost of the cloud path is set by the provider rather than by the app.
Writing modes, empowers, a dictionary and snippets
The shaping layer is where this differs from ordinary dictation, and it is split across four features.
Writing modes define what happens to a transcript. You set instructions for messages, emails, notes and tasks, and enable enhancement, shortening or translation. The instruction editor screenshot shows example Russian instructions to remove fillers, preserve tone, add paragraphs, and avoid inventing facts or promises. That last clause is telling: the mode is a prompt, and the prompt tells the text model not to make commitments you did not make. The README notes the diagram in that section uses illustrative text examples rather than model benchmark results, which is the right kind of caveat.
Empowers are for one dictation rather than a category. You append your own instruction to a single dictation, choose a colour and preview the recorder effect, so the thing you are saying gets a different treatment for that one time.
The personal dictionary holds names, terms and your own corrections, which is the mechanism for fixing the recurring recognition errors that make a user give up on dictation. Snippets hold frequently used text fragments at hand.
The history is more interesting than it sounds. You can revisit past dictations, and audio is retained for the ten latest recognition failures without text so you can retry them. Keeping the audio of failed attempts only, and only ten of them, is a deliberate middle position: enough to recover a dictation the recogniser mangled, without keeping a voice archive. The insights view then reports word counts, speaking speed, activity and dictation usage across applications, with the note that the screenshots show demo data and that time saved is an estimate.
The interface is available in English and Russian, or you can follow your system language.
A recorder you can pet, placed along the top edge
The recorder is the visible part of the application, and the README treats its appearance as a feature list of its own.
You pick waves or a pixel companion, adjust the size and choose a palette, and the robot reacts when you pet it with your pointer. The recorded appearances are named in the asset descriptions: a lavender Prism, a green Aurora, a pixel companion, and a heart in response to petting.
Placement is the other decision. The recorder can sit in a floating bubble or along the top edge of your screen, and the top-edge mode has platform-specific behaviour: on a Mac with a notch the robot sits beside the cutout, and on Windows it sits in the centre. There is a floating-recorder size control shown in the animation.
It would be easy to read this as decoration, and it is, but decoration in a tool you use dozens of times a day is not the same as decoration in a screenshot. A visible indicator of recording state is what lets you keep your eyes on the screen you are working in rather than on the app, and the petting behaviour is a small reward for a tool that otherwise interrupts you.
The practical detail is that this is a floating overlay, which means it interacts with the window manager and with whatever else is on screen. How the top-edge recorder behaves across multiple monitors is not documented in the README, so that is worth checking on your own machine before you settle on a placement.
Version 0.1.12, and the source that was opened in it
The release notes in the README describe 0.1.12, and one line in them matters more than the rest.
This update adds a recorder preview state wheel, refined surfaces and transitions, a Graphite palette and simpler hidden-page navigation. And then: the original source and documentation are now available under Apache License 2.0. Updates preserve your library, preferences and saved credentials.
That is a change in kind for the project. The repository is named no-tipe-releases, and the fact that the application source and documentation are now published under Apache-2.0 means the earlier releases can be examined and the project can be built, audited and contributed to. The top level of the repository reflects that layout: NoTipe/ for the application itself, CrossPlatform/ for the platform specific parts, DesignAssets/ for the visual work, Scripts/ for build and packaging scripts, docs/ for the documentation, and a licenses/ directory alongside LICENSE, NOTICE and THIRD-PARTY-NOTICES.md.
Two files describe the origin of the code. SOURCE.md and SOURCE-SNAPSHOT.json sit at the root next to BUILDING.md, which is where the build instructions live. A source snapshot record alongside a build guide is the shape you would expect from a project that has published the code of an application which was previously delivered only as binaries.
The releases are frequent and recent: v0.1.10 on 2026-10-06, v0.1.11 on 2026-10-10 and v0.1.12 on 2026-10-13, with the last push to the repository on the same day as the last release. The README carries English and Russian versions, and the website is notype.tech with the downloads linked from the GitHub releases.
Where No Type is the wrong tool
There are cases where a dedicated dictation application is the wrong answer, and some where this one specifically does not fit.
On an Intel Mac, No Type does not run at all. The download table states the requirements as macOS 13 or later and Apple Silicon, and the Windows build is x64. There is no Intel build listed, which is a hard exclusion rather than a preference.
On macOS without Accessibility permission, the app copies to the clipboard instead of inserting, which turns a one action flow into two. If your workflow depends on text landing in the field without a paste, that permission is a prerequisite rather than a nicety.
The privacy question has to be answered per configuration rather than per product. A local speech model with a local text model keeps the audio and the transcript on the machine. A cloud provider with your own key sends both off the machine and bills you for it. And the app keeps the audio of the ten most recent failed recognitions, which is a small archive of your voice that exists by design.
The version line is still 0.1.x, and the interface is English or Russian only.
The alternative is the dictation built into the operating system. The difference in approach is worth stating without pretending the comparison is close. Built-in dictation is tied to the operating system's own recognition. No Type costs nothing, behaves the same on both platforms, adds the writing modes and the dictionary that shape the transcript after recognition, and gives you the choice of running the recognition locally. If you only need raw speech to text, the system tool is fewer moving parts. If you need the transcript rewritten before it arrives, or the recognition to stay on your machine, that is what this application is for.
Editorial conclusion
No Type suits somebody who writes long messages, email and notes by voice and wants the transcript cleaned up before it lands in the field, and it suits people who need the recognition to run on their own machine rather than on someone else's servers. It does not suit a Mac with Intel silicon, since the macOS build requires Apple Silicon and macOS 13 or later, and it does not suit a workflow that needs a cloud speech provider included in the price, because the providers are billed separately under their own pricing. Check three things before you install it: that you grant Accessibility permission on macOS if you want text inserted rather than copied to the clipboard, which the README is explicit about, which speech model and which writing-mode text model you will use, since one of them can be local and the other a cloud API, and whether the failure-audio retention setting matters to you, since the audio of the ten most recent recognitions without text is kept so you can retry them. The last push to the repository was on 2026-10-13 and the current release is v0.1.12.
Frequently asked questions
What does No Type do and which platforms does it support?
It turns speech into text for your apps, putting the dictated text into the focused field. It ships for macOS 13 or later on Apple Silicon and for 64-bit Windows, and the source and documentation are published under Apache License 2.0.
How do I trigger dictation in No Type?
You hold the shortcut while speaking, or press it to start and stop, and both the shortcut and the recording behaviour are configurable in settings. macOS uses Option by default, Windows defaults to Right Ctrl, where holding it while speaking and releasing it finishes the recording, and updates keep your saved shortcut.
Can No Type run speech recognition offline?
Yes. You can download a speech model to your computer, and writing modes can process the transcript with a local text model as well. The README lists offline among the project topics, while a cloud provider connected with your own API key is the other option and is billed separately.
What are writing modes in No Type?
A writing mode defines what happens to a transcript. You set instructions for messages, emails, notes and tasks, and enable enhancement, shortening or translation, then choose a local or cloud text model to apply them. The example instructions in the README cover removing fillers, preserving tone, adding paragraphs and avoiding invented facts or promises.
Does No Type need Accessibility permission on macOS?
For text insertion, yes. The README says text insertion needs Accessibility permission and that without it the result is copied to the clipboard instead, so the flow becomes a paste rather than direct insertion.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/darlingfxx02-no-tipe-releases)Was this useful?