AI dictation software for legal documents
Stop typing first drafts. Talk the section through.
Talk a contract clause, a motion section or a discovery response through, then tell the AI agent what to make of it: headings, numbered points, a summary. Pick a local model and it's drafted on your own machine, free and unlimited on every plan.
Open source · 80,000+ downloads
Why dictating a contract stalls
Talking it through was never the slow part. Shaping it is.
Speak an engagement letter or a case chronology and you'll get every word back in one unbroken block. Cutting that block into clauses, headings and numbered points is what eats the afternoon, and plain dictation hands that part straight back to you.
One block of speech, not a clause in it.
Dictation cleanup hands back what you said, tidied. It will break the speech into paragraphs and number a list you clearly dictated. What it won't do is impose a heading or restructure a deposition summary into sections. That job belongs to the AI agent, and you have to ask it for it.
The motion is the sensitive part, not the audio.
A transcript is words. A motion is the argument you're about to run and the position your client will take. Choose local processing when you want the composing to run on your machine, or cloud tools when they fit your workflow.
Paralegals and legal ops draft it too. Per-seat pricing adds up.
An associate dictates the contract, a paralegal checks the cites, an assistant cleans it up, legal ops files it. Charge per seat and the cost follows the draft across four desks. OpenWhispr charges nothing for a seat on Free, so each of them can use it.
Party names and statutes come back mangled.
A general-purpose model has never seen your matter. Put the case names, the parties, the Latin and the citations your practice uses in the custom dictionary. A dictation vocabulary only fixes what was heard; this one also reaches the model that drafts, so it sees those names before it writes a word. It moves the odds rather than guaranteeing them, so read what carries legal weight before the document goes out.
From microphone to document
You dictate a section. You get back something worth editing.
A transcript isn't a document. Dictation, cleanup and the AI agent each hand you something different.
Say it once, read it back punctuated
Talk an interrogatory response through the way you'd say it out loud, and it comes back transcribed and punctuated with the filler and the false starts taken out. Address the agent instead and it works from the same dictation, handling the disfluencies itself.
Headings and numbered points, not a wall of speech
Cleanup breaks your speech into paragraphs and numbers a list you clearly dictated, and stops there. Ask the AI agent instead, on its own hotkey or by name mid-dictation, and it composes what you asked for: the headings, the numbered points and the lists you talked through.
Nothing to install into Word or your DMS
One global hotkey copies the text and pastes it at your cursor, so the clause lands in Word, a matter record, an email or a web form the same way. Auto-paste is on by default, so you edit it where it lands.
Audio kept on your own machine by default
30 days
Audio Retention ships on at 30 days. Settings takes it from 1 day to 90, or switches it off
Languages you can dictate a matter in
100+
a German framework agreement comes back as German text
Cost to add the paralegal and legal ops
$0/month
no card details, and no per-seat licence to buy
How a contract gets drafted
Talk through the motion, then edit what comes back.
Three steps from talking the argument through to editing it in the document.
Install it and choose where it runs.
Setup puts a local model and OpenWhispr Cloud in front of you, with Cloud badged Recommended and local warning you it takes a few extra steps. Going local means pulling down a speech model of 75 MB to 3 GB and a separate local LLM of 0.25 GB to 19.8 GB first, depending which two you pick.
Talk the clause through, ums and all.
Say the deposition summary the way you'd explain it to a colleague, ums and all. Cleanup hands the same words back with the filler and the false starts stripped out.
Ask the agent for headings and numbered points.
Highlight the rough passage, press the agent hotkey and tell it what you want, and it rewrites that passage in place in the document. With nothing highlighted the draft arrives in the app's assistant panel instead, ready to copy across. Either way it's plain markdown text rather than Word styles, so you shape it in the document itself.
Ten claims you can check
What OpenWhispr does to a legal document, and where it stops.
What it does, what it won't do, and where the document ends up. No asterisks.
The setting that would upload a contract arrives switched off
Pick a local model at setup and the audio, the transcript and the composed section all stay on the machine you spoke into. Two deliberate acts send content to us: switching on paid cloud sync, which arrives off and stores your notes, transcripts, agent conversations and your custom dictionary, party names included; and sharing a note or moving it into a team space.
Cleanup keeps your wording and won't add a sentence you didn't say
Cleanup is told to keep your voice, your wording and your intent, and never to add a sentence you didn't say. It still fixes grammar and punctuation, splits run-on sentences, drops false starts, and writes spoken numbers and dates out in standard form. That's an instruction to a model rather than a guarantee, so read an indemnity back before it goes out.
The agent turns talk into a numbered section
Say its name mid-dictation or press its hotkey, tell it what you want, and it composes that from what you talked through: headings, numbered points, a summary of the argument. Highlight a passage first and it rewrites that passage in place; with nothing highlighted the draft lands in the assistant panel. Pick local at setup and it composes on the LLM you downloaded, on your own machine. Styling stays a Word job.
One hotkey, and no add-in for IT to approve
The text goes to the clipboard and pastes itself where your cursor is. On macOS, without Accessibility permission it can copy for you but not paste.
Know where the ceiling is before you hit it
For selection edits, select 6,000 characters or fewer.
Dictate it into a note now, shape it later
Save the case chronology as a note while it's fresh, and a note action turns it into clean markdown when you come back to it.
Recordings are files, so deleting one really deletes it
Recordings are audio files on your machine you can open or delete yourself; notes and transcripts sit in a local database the app deletes rows from. Audio Retention ships on at 30 days, and Settings takes it from 1 day to 90 or switches it off. Transcripts are the other way round, kept until you set a window of your own, so set one alongside the audio. OpenWhispr doesn't encrypt that local data itself, so the folder wants the care you'd give any other matter folder.
Send procurement the repo instead of a policy PDF
OpenWhispr is open source. When IT asks what happens to a contract you dictated, they can inspect the code rather than take our word for it.
The paralegal and the legal ops lead get the same hotkey
Legal assistants and legal ops draft these documents alongside lawyers. Free is a download rather than a licence, so the person who assembles the interrogatory response has the same hotkey as the person who signs it.
Dictate the deal in the language it was done in
Whisper-class models cover 100+ languages, so the deal you did in Spanish comes back transcribed in Spanish.
Before the first document
What actually comes back when you dictate a contract.
Anything we've missed, bring it to our Discord, where the people who build OpenWhispr answer.
Which legal documents can AI dictation draft: a contract clause, a motion section, a discovery response?
Who else ends up with a copy of an unsigned contract?
Can it write a whole brief in one go?
Will party names and citations survive a deposition summary?
Can a paralegal take over the contract after I've dictated it?
What does it cost to run a matter's documents through it?
Straight from our users
Rambling in, editable text out.
Nobody was asked for these. One of them stumbles mid-sentence and still gets a clean paragraph back, which is what dictating a clause actually sounds like.
0
Client documents sitting on our servers
In local mode, signed out, with cloud sync off and nothing shared, nothing reaches us. Turn cloud sync on, a paid setting that arrives switched off, and your notes, dictation transcripts, agent conversations and custom dictionary are stored in our cloud database.
Started using OpenWhispr with my local models and hardware. I have to say, this is better than typing.
I switched to voice input for my Claude Code chats. Even when I stumble and rephrase mid-sentence, the transcription comes out clean and polished.
I needed a local model for when I'm travelling or have no WiFi, and this perfectly solves it. The UI and performance are great.
Pricing
Free to get started.
Upgrade when you want more.
FreePopular $0No card details needed | Pro $6.67/user/mo$80/user billed annually Start with Pro | Business $13.33/user/mo$160/user billed annually Start with Business | |
|---|---|---|---|
| Best for | Paralegals and legal teams drafting contracts, briefs and motions, who don't mind downloading a speech model and an LLM first | Legal assistants who'd rather dictate through OpenWhispr Cloud than manage model downloads, at the desk and on an iPhone | Legal ops teams handling document work across a lot of matters, and recording calls all month |
| Main reason to choose | Free plan: $0. | The cloud speech-to-text meter comes off, and device sync and the iPhone app come with it | Cloud meeting recordings go unlimited and team features unlock, at $16 per user per month |
| Offline speech-to-textOffline speech-to-text runs entirely on your own device, it works without internet connection, and it's unlimited and free. | Included | Included | Included |
| Cloud speech-to-textCloud speech-to-text sends audio to OpenWhispr Cloud and does not require local models to download. | 2,000 words/week via OpenWhispr Cloud, or unlimited using your own API keys | Unlimited OpenWhispr Cloud speech-to-text | Unlimited OpenWhispr Cloud speech-to-text |
| Meeting recordings | 5 hours/month | 20 hours/month | Unlimited |
| Privacy & data handling | Offline audio stays on your device. Cloud audio is sent only to transcribe it, then never stored. | Offline audio stays on your device. Cloud audio is sent only to transcribe it, then never stored. | Offline audio stays on your device. Cloud audio is sent only to transcribe it, then never stored. |
| Setup experience | Hands-on: install app, choose local models, optionally add API keys | Easy: install app, upgrade, use unlimited cloud speech-to-text | Moderate: configure users, Agent mode, and data workflows |
| Advanced features |
|
|
|
| Team management | None | None | Basic team / workflow use |
| Support | Community support | Email support | Priority support |
Need SSO, compliance features, admin controls, or dedicated support for a larger team?
We also offer a custom Enterprise plan with managed onboarding and tailored features.
Talk to us
Read the code yourself
A contract shouldn't need a round trip to get written.
Turning a transcript into clauses and numbered points takes an AI model. OpenWhispr's AI agent can do that work with a model you downloaded yourself, served by the llama.cpp binary that ships inside the app, on 127.0.0.1.
You don't have to take our word for that. OpenWhispr is open source, so your risk team can inspect the code rather than read a policy about it.
Break something at 11pm the night before a filing deadline and you're talking to the engineers who wrote it, in Discord.
$0 a seat · No card details
Say the clause out loud. Get a numbered section back.
Download it free, pick a local model at setup, then hand the AI agent your rough dictation. Highlight the passage and it rewrites the clause or the case memo in place, as plain markdown text. No card details, no seat to buy.
Loading...Available on macOS, Windows, and Linux. Get community support in Discord.