- Runs on
- Your Mac, on Apple Silicon
- Needs
- A one-time engine install and a 2.5 GB download
Free, open-source dictationmacOS, Windows, Linux
Learns your words.
Knows when not to use them.
Dictate into any app with free local speech models. Hold a shortcut, speak, and Entune pastes the text where you are.
uv tool install entune && entune
Needs uv, which installs Python for you. Or pip with Python 3.12+.
What it does for you
You already dictate. This is what changes.
-
01
Corrections that fit
Your dictionary fixes names and terms. When one could also be an everyday word, it picks what the sentence means, and never rewrites you.
-
02
Fast, even when you talk long
Fast mode transcribes at natural pauses while you speak, so when you stop, only the last part is left.
-
03
Clean text, nothing rewritten
Ums go, paragraphs and lists appear. Your words stay yours. Each is a switch you turn on.
-
04
A dictionary from day one
Learn from recordings you already have. Entune proposes your names and terms, and you review them.
-
05
Cloud or local, one page
Paste a key for a cloud model or download a local one, all from the same page.
-
06
Retry with another model
Every recording is kept. Transcribe it again with another model, or retry when a provider fails.
-
07
Drop any recording
Drop audio files onto the window and they are transcribed into History, like a dictation.
-
08
Anonymous mode
One switch blurs your History for screen sharing. The newest stays readable until you copy it.
-
09
MCP integration
Agents build and refine your dictionary. They check how a word appears in your transcripts and change entries only when you agree.
Corrections that consider the context
A match is not always a replacement.
After transcription, when a dictionary match could mean your term or an everyday word, a decision model reads the words around it and applies only the meaning that fits. It chooses from your own entries. It never generates or rewrites your text.
- Jev
- Fewest wrong changes in our test. Sends the matched context to TypeSafe, on your own key.
- OpenAI
- OpenAI's Decisions API. The context goes to OpenAI, on your own API key.
- Perplexity
- Caught the most fixes in our test. The context goes to Perplexity, on your own API key.
- Laya
- Runs on your computer, in English, after a one-time install.
Ask cloud Claude to review this.
Replaced: the name fits here
Move the files to cloud storage.
Move the files to cloud Claude storage.
Kept: the everyday word is right
Broken: find-and-replace changed every match
Same entry, two sentences. Find-and-replace would break one of them.
A test on real dictation
Which model knows when to fix a misheard word?
- places the dictionary offered a word
- 121
- should change
- 47
- should stay
- 72
- PerplexityDecisions API Fixes caught: 43 Wrong swaps: 3
- JevFewest wrong swaps Fixes caught: 38 Wrong swaps: 1
- OpenAIDecisions API Fixes caught: 42 Wrong swaps: 8
- LayaRunning locally on a Mac Fixes caught: 28 Wrong swaps: 8
- Replace every matchHypothetical baseline Fixes caught: 47 Wrong swaps: 72
In this test, Perplexity caught the most fixes. Jev made the fewest wrong swaps.
- Dictionary learned from 7.5 hours of the maintainer's dictation, transcribed by Parakeet, with entries suggested by GPT-6 Astra on a ChatGPT plan.
- Tested on the next 2.1 hours it had never seen: 160 recordings.
- Expected answers written down before any model ran. Jev, OpenAI and Laya were tested on October 7; Perplexity on October 8 with the same transcripts, dictionary and expected answers.
- One speaker, one reviewer judging from text, one run per model.
Two matches could not be judged and are excluded; Perplexity changed one of them. Each bar is a share of its own total: fixes out of the 47 places that should change, wrong swaps out of the 72 that should stay. These results do not establish a statistical ranking, including between 1 and 3 wrong swaps, or measure overall transcription accuracy.
In the separate formatting test, of 92 places judged to need a paragraph or bullet, Jev placed 35 with 12 unwanted, Perplexity 32 with 5, OpenAI 8 with 1, and Laya 5 with 37.
Data, method and limits →Your models
Three jobs. You choose the model for each.
Entune is free and MIT licensed, with no Entune subscription. Local speech models run on your computer without an API key or per-minute payment. Cloud speech, cloud decision models and dictionary generation use your own accounts and can cost money.
- Runs on
- TypeSafe
- Needs
- Your TypeSafe key
Parakeet keeps your audio here. Jev gets only the text it decides on.
One provider key for dictation, billed by that provider. Building the dictionary uses your ChatGPT plan, only when you ask. Entune has no subscription and sells no credits.
Our starting points. Compare any two on your own voice: in History, transcribe a recording again with another model.
- Runs on
- OpenAI, on your ChatGPT plan
- Needs
- Nothing to set up. Sign in and approve.
A dictionary that learns
Teach your speech model your words.
Local models such as Parakeet and Whisper.cpp are fast and keep your audio on your computer, but they can mishear names and jargon. Entune closes that gap with recordings you already have: it transcribes them with the model you dictate with, and builds a dictionary from the words that model gets wrong.
Entries belong to the speech model they were learned on, because every model mishears differently. Pin an entry to use it with every model; suggestions can't remove pinned entries. Later runs can revise or remove learned entries, and you review each change.
History you keep
Every recording, on your disk, with what happened to it.
Replay the audio. Copy the text. See what each processing step changed. Transcribe a recording again with another model and compare. Provider failures stay visible. Nothing is uploaded to Entune, because there is no Entune server.
So um, um, we ship on Friday. Three things for today. - The build is green. - Docs are updated. Ask Claude to review the notes.
All three start off, in Settings › Corrections & formatting, and each needs a decision model. The original transcript is always kept in History.
Privacy, said plainly
No account. No telemetry. No hosted history.
Speech. Cloud speech sends audio to the provider you chose. Local speech keeps it on your computer.
Corrections and formatting. Off until you turn them on. Then your decision model reads the text: Jev sends it to TypeSafe, OpenAI's Decisions API to OpenAI, and Perplexity to Perplexity, even when speech is local. Laya keeps it on your computer.
Your dictionary. Building it runs only when you ask, and sends the transcripts it reads, and the entries in them, to the language model you chose.
Your files. Audio, transcripts, dictionary, settings and keys live in one folder. Settings › Data & privacy exports your audio and transcripts, or deletes everything.
Use Entune on a computer only you use: its local API has no password, so other programs on it can read your recordings. Data and privacy, in full
- 01MicrophoneHold the shortcut, speakStays here
- 02Speech model · ParakeetOn your computerStays here
- 03Your dictionaryEntries that match the transcriptStays here
- 04Decision model · LayaOn your computerStays here
- 05Pasted where you were typingSaved to Entune's folderStays here
Local speech and Laya: dictation stays on your computer. With corrections, filler removal and formatting all off, nothing goes to a decision model.
Questions
Every answer, without the wall of text.
Each topic opens to the detail you need, then links to the full guide. New to dictation? Start with what dictation software does.
Is it really free?
Yes. Entune is MIT licensed, with no Entune subscription. Local speech models run on your computer with no API key or per-minute payment. Cloud models use your own accounts and can cost money.
Choosing models →Does my audio leave my computer?
Only if you choose cloud speech, and then only to the provider you picked. Local speech keeps it on your computer. There is no Entune server.
Dictate offline →What does the decision model actually do?
When a dictionary match could mean your term or an everyday word, it reads the words around it and applies only the meaning that fits. It chooses from your own entries. It never generates or rewrites your text.
Decision-model results →Do I need an API key?
Not for local speech. To build your dictionary, choose Sign in with ChatGPT and your plan's usage applies. Prefer a key? OpenAI, Anthropic, Google Gemini, Groq and Mistral work too.
Correct names and technical terms →How long does building a dictionary take?
It depends on how much you give it. In our October test, 7.5 hours of dictation took 39 minutes in 10 parts. Dictation keeps working throughout, and Stop keeps finished parts for Continue.
Dictionary reference →Which systems does it run on?
macOS, Windows and Linux. macOS has been tested by hand; Windows and Linux are covered by automated tests. Linux is new, with shortcuts and paste on X11 and Wayland.
Using Entune →Can I dictate prompts to a coding agent?
Yes. Entune pastes into the input you are in, such as an agent's chat box in your editor or a prompt in your browser, and your dictionary keeps library names and flags right. On a Mac, delivery into a terminal has been tested in Ghostty. There is a guide for Claude Code too.
Dictate to coding agents →Install
One command. Then three steps.
uv tool install entune && entune
Puts Entune in Applications and opens it. The first start can take up to a minute.Puts Entune in the Start menu and opens it, in Windows PowerShell.Puts Entune in your applications menu and opens it. Linux needs a few system libraries; see the Linux guide. Needs uv; without it, on Python 3.12 or newer: python -m pip install entune && entune
-
01
Set up a speech model
In Models, paste a provider's API key or download a local model. The first one becomes your default. Parakeet, on Apple Silicon, first needs its engine:
uv tool install parakeet-mlx. -
02
Allow permissions
On macOS: Microphone, Accessibility and Input Monitoring, each from its own button.On Windows, only the microphone.On Linux, keyboard access: one command, run once, then log out and back in.
-
03
Set a shortcut
Hold it in any app and speak, or toggle hands-free recording. The text is pasted where you are, or copied if no text field is active.
You can start without a dictionary or decision model and add them later. To upgrade: uv tool upgrade entune && entune