Typort

One app instead of four subscriptions

Dictation, screen recording and meeting memory normally cost ~$50/user/month across cloud tools that don't talk to each other. Typort does it all - on your own PC for free, or on the web from any machine. And local mode doesn't even need a GPU - most CPUs can run it, free forever. NVIDIA's Parakeet engine flies on the chip you already have.

CLOUD MCP · CONTEXT ENGINE

Fully searchable. Connected to your favorite AI.

CONNECT

Connect your AIs with a shared memory

ChatGPT, Claude, Claude Code, Gemini, Perplexity and Grok share one searchable memory, so they can collaborate. Ask any of them what any of them said.

SPEAK

Dictation anywhere

Speak. It types - into any Windows app at speaking speed. Keyboards do ~45 words a minute. You talk at 150.

SHOW

Record your screen, share a link

Record your screen and send the link to anyone. It's live before you even stop recording - whoever you send it to can start watching while you're still talking. Every 3 seconds is already saved, so a crash can't eat your video.

MEET

Meetings that file themselves

Say "Typort meeting" - it records, transcribes, files the transcript under the right project and pops insights every two minutes. Every meeting lands in your vault: completely searchable, and available to all your AIs.

How each piece works

Dictation

Hold Ctrl + Win (or F9 in terminals)

Hold, speak, release: text lands at your cursor in any app. Terminals, IDEs, browsers, Word, games chat.
Runs NVIDIA's Parakeet model right on your CPU - no GPU needed. The model is loaded before you ever press the key, so the first dictation is as fast as the hundredth.
It learns you: every correction you make is banked as training data for your own personal speech model.
Two lanes, same app: free forever on your own machine - plain CPU, no GPU needed (local lane is English-only for now), or the web tier runs it in the cloud with more languages - your work laptop, any machine. Not a fallback. Your layer, everywhere.
Voice commands, all prefixed "Typort": "Typort polish" rewrites your draft, "Typort screenshot", "Typort record", "Typort meeting", "Typort compose", "Typort fact check" and more.

Screen recording

Say "Typort record" - again to stop

The share link works while you're still recording. Send it mid-take; the viewer watches live and it becomes the recording.
Crash-proof: footage uploads in 3-second pieces as you record. Power cut at minute 14? You have 13:57.
Keep recordings local-only with one toggle.

Meeting mode

Say "Typort meeting" - again to end

Records in 30-second chunks, transcribes each locally, keeps going for hours.
Every 2 minutes an AI summary pops up: key points, decisions, action items.
It sees the screen, not just the audio: when something relevant appears - a slide, a document, a demo - it detects the change and grabs a screenshot at that moment. You can't feed an hour of video to an AI, but you can feed it the dozen screenshots that mattered. The insights are built on both.
Say "Typort meeting" again and the full transcript is assembled and filed under the project you were working in, automatically.
Crash recovery is automatic: if anything dies mid-meeting, the transcript rebuilds from the chunks on restart. You lose 30 seconds at most.
Records your mic or the other side of the call (system audio), your choice.

The Vault - memory + MCP

Desktop app + free Chrome extension

Continuously captures your AI conversations right in the browser - ChatGPT, Claude, Grok, Gemini, Perplexity - plus Claude Code and Slack, into one searchable vault.
It's an MCP server: plug your history into Claude, ChatGPT, or anything that speaks MCP - any agent, any AI app - and just ask: "What did John say in last month's meeting about the budget?"
Fully searchable - by meaning, by date, by project.
Cloud notetakers have MCP now too - but it serves their copy of your meetings, from their cloud, rationed by their plan tiers. Typort gives you the choice: keep it on your own disk where it never leaves your machine, or go Pro and get the full cloud experience everywhere. Private when you want private, everywhere when you want everywhere.

Get the free Chrome extension →

Personal AI Model Training

Nobody else offers this. Not the dictation apps, not the big platforms - nobody.

Every time you fix what it typed, Typort banks the audio and your correction - and fine-tunes NVIDIA's Parakeet speech model on your speech patterns. Not a dictionary - a real model trained on how you talk.

Pro feature - or 100% free for people with speech disabilities. Get verified →

Accents

Australian, Scottish, Indian, Southern US, Korean-English - your personal model handles them all.

Speech Disabilities

Off-the-shelf models often fail anyone whose speech doesn't fit the mold - around 60% accuracy. A model trained on YOUR voice: 95%+.

Technical Jargon

Your industry terms, product names, acronyms. Say it once, correct it once, learned forever.

Speech Disability Access

100% free for people with speech disabilities. Local AND cloud. Forever. No expiration. Get verified →

Your trained model is portable. It's a file. Take it to any machine. It's yours even if Typort disappears tomorrow. With Pro, we host it for you - so your model follows you to any device.

	Typort	Wispr Flow	Loom	Granola
Windows native	Yes	Yes	Yes	Yes
Audio stays on your machine	Yes	No - cloud only	No	No - cloud
Dictation into any app	Yes	Yes	No	No
Screen recording + share links	Yes	No	Yes	No
Meeting notes + AI insights	Yes	20-min sessions	Paid AI tier	Yes
Ask AI about your history (MCP)	Yes, with Pro - any AI queries your vault	No	No	Cloud, $14/user/mo
Price	Free · Pro $10/mo	$15/mo	$18-24/mo	$14-35/user/mo

Pricing

Local processing costs us nothing, so local is free. The cloud parts are what you pay for.

Free

Download and use forever

Dictation, meeting mode, the Vault - everything runs locally on your machine
Every correction you make builds your personal voice model - your training data stays on your machine
English dictation, tuned hard for accuracy - more languages coming
Free Chrome extension included
Community support only.

Download free

Pro

~~$20~~ $10/mo

or ~~$200~~ $100/year • 870 of 1,000 founder spots left

Your own trained model - trained to YOU in our cloud. Your corrections become a model that knows your accent, your names, your jargon. Nobody else's voice, just yours. English only for now.
Coming soon to Pro accounts: iPhone and Android access - voice typing on your phone, powered by your own trained model
Cross-device sync and cloud backup of your vault
Everything fully searchable - and any AI can query your vault over MCP. Granola charges $14/user/mo for this.
Priority support

Get Pro

Enterprise

Custom

Per-seat, annual

Team AI governance: audit trail, PII detection, policy enforcement
SSO and admin dashboard
On-prem / air-gap deployment
Dedicated support

Typort for Business →

Free for people with speech disabilities - local and cloud, forever, no expiration. Get verified.

FAQ

Does my audio ever leave my machine?

Not on the local lane, period. Dictation, meeting transcription and the vault all run on your own hardware - verify it yourself with Wireshark: zero network calls during local dictation. The cloud lane is a feature you choose, not a fallback: turn it on when you want Typort at work, on a locked-down work laptop, or - soon - from your phone. Want privacy? You've got privacy. Want it everywhere? You've got that too. And that choice is the point: install a cloud notetaker and your meetings live on someone else's infrastructure - their terms, their acquisitions, their breaches. Files on your own disk have none of that surface.

Do I need an NVIDIA GPU?

No - there's no GPU in the picture at all. Local dictation runs NVIDIA's Parakeet model on your CPU, fast on any modern machine. The $10/mo web tier runs everything in the cloud on any machine - phone apps on the way. Same app, pick per device.

Does it work in Cursor, terminals, games?

Yes. Typort simulates keyboard input, so it works in anything that accepts text. F9 exists as an alternate hotkey precisely because terminals eat Ctrl+Space.

How is it free? What does the $10 buy?

Local processing runs on your hardware, so it costs us nothing per user - that's why it's free and stays free. Pro pays for the parts that cost real server money: hosting your personal voice model so it works on any device, cloud sync, backup and support.

Typort vs Wispr Flow?

Wispr Flow is excellent dictation, but transcription always happens in their cloud and costs $15/mo forever. Typort's dictation runs on your own machine for free, and it also records your screen and remembers your meetings - two more subscriptions you don't need.

Typort vs Granola?

Granola makes great meeting notes in their cloud. Typort's meeting mode transcribes locally and files transcripts by project on your own disk - and with Pro, any AI can query your vault over MCP. That's the feature Granola charges $14/user/mo for.

Typort vs Dragon?

Dragon is an 8+ GB install from the pre-AI era. Typort is a small app that runs NVIDIA's modern Parakeet model on your CPU, trains itself on your corrections, and doesn't cost $699.

Is my meeting data private?

Recorded, transcribed and stored on YOUR machine. This is why lawyers and healthcare workers use local tools - several bar associations now require that AI notetaker data be retained on your own device.

What do I need to run it?

Windows 10/11 (64-bit), 8 GB RAM, ~1 GB disk including the speech model. No GPU needed - transcription is instant on any modern CPU. Mac version coming soon.

Why is the app so small?

It's built in Rust, not Electron. No bundled browser runtime, no 400 MB of Chromium - the whole app is ~65 MB and idles at ~50 MB of RAM. Microsoft, Google, Amazon and the Linux kernel are all adopting Rust for the same reason: speed with memory safety.

Refunds?

14 days on Pro, no questions asked.

Are screen recording share links private?

Share links are unlisted URLs - anyone with the exact link can view, nobody can browse or search for them. Password protection is coming. You can also keep recordings local-only with one toggle.