Voice to text, wherever you want it

Hold a key, talk, and the words land wherever you're typing. Free + open-source.

Tap or hold

right

to blurt

Hey Sam, did you see Assembly's new Blurt app? I actually used it to send you this Slack.

How it works

You ramble. Blurt writes it clean.
Write in any style.
Type 4x’s faster.

Uh, I don't know what to say, ummm hmmm, and I need to think about this fully. So what I meant to say is that you're right.

I don't know what to say, and I need to think about this fully. What I mean to say is that you're right.

Filler words removed

Keyboard

45 wpm

I’m starting a new project called, Elia Launch, and I need a project manager on the project asap. Is that possible to do now?

Blurt

220 wpm

I’m starting a new project called, Elia Launch, and I need a project manager on the project asap. Is that possible to do now?

Why Blurt

Dictation you can actually trust.

Built for accuracy

Up to 30% fewer hallucinations than Whisper, and clean text back in under a second — powered by the AssemblyAI Dictation API.

Thoughts to text in milliseconds

Transcription lands in about 150 ms, cleanup in under a second. No local model to load, no job queue to poll.

19 languages

Dictate in 19 languages and code-switch mid-sentence. The model follows you between them — no setting to flip.

Free and open source

No packages you’ve never heard of, no supply chain to audit. MIT licensed — fork it and make it yours.

Better than built-in

Think macOS dictation, but fast, accurate, and working everywhere you can type — powered by the AssemblyAI Dictation API.

Your voice, your style

One style for Slack, another for email. Define them in a prompt, switch between them from the main menu.

AssemblyAI Dictation API

Want to build your own Blurt?

Everything that makes Blurt fast and accurate is AssemblyAI's Dictation API. The same Dictation API is yours to build on.

Get an API key Learn more

import assemblyai as aai

aai.settings.api_key = "<YOUR_API_KEY>"

result = aai.DictationTranscriber().transcribe_live("clip.wav")

print(result.text)  # verbatim transcript
print(result.llm_response)  # cleaned-up text, ready to send
print(result.final_text)  # the cleaned-up text, falling back to the transcript

Common questions

: Blurt is an open source dictation app that turns speech into text anywhere you type. It runs on AssemblyAI's Dictation API, the same transcription, punctuation, and formatting pipeline behind production voice products. Fork it, change it, ship it as your own.
: Blurt is free and open source under the MIT license. You bring your own AssemblyAI API key and pay for what you transcribe, around $0.62 per hour of audio. New accounts start with free credits, so you can dictate for a while before you spend anything.
: Text comes back in about 134ms at p50, though your numbers shift with audio duration.
: Blurt runs on AssemblyAI's Universal speech model, which holds up on accents, code, and domain jargon that general dictation tools drop. Add your product names, acronyms, and function names through custom vocabulary and they come out spelled right the first time.
: Blurt runs on macOS 14 and later. Press the hotkey, talk, release, and the text lands in whatever app has focus: your editor, your terminal, Slack, a doc.
: Start with the Blurt repo. It is a working reference for the parts that are tedious to get right: audio capture, streaming to the Dictation API over WebSocket, partial transcript handling, and text injection into the focused app. Swap the UI, keep the pipeline, ship your own product.
: Wispr Flow is a commercial app you subscribe to. Blurt is source you own. If you want dictation that works the moment you install it, Flow is a good product. If you want to change how dictation behaves or ship it inside something else, start from Blurt.