VantoLabsVantoLabs
Beta on macOS · Windows on the way

You speak. The text lands where it belongs.

We are VantoLabs. We build a desktop voice assistant. One keyboard shortcut turns speech into text in any application, and your voice notes and meeting summaries land in your Obsidian vault as plain Markdown files. Everything is computed on your own machine, and Polish is the first language here rather than an add-on.

We build this with ADHD in mind, because speaking is the shortest path from a thought to a note.

Voice note
recording…

you speak — VantoLabs writes

Everything stays on your computer.

Transcription, summaries and note organisation all run locally. You create no account and send no recordings to any server.

Press the shortcut

Cmd/CtrlShiftSpace

in any application

Speak

dictate a thought, a note, or record a meeting

The text is in place

clean text in the active field, a note in Obsidian, or a meeting summary with an action list

Product

What VantoLabs Voice Assistant does

Three recording modes. In each of them you speak, and organised text lands where you need it.

Dictation
Press the shortcut and speak, and the text is pasted into the active field of any application. The model punctuates and writes numbers as digits. Your own glossary holds the names, companies and industry terms you do not want mangled.
Buy the Thursday morning tickets…
Voice note
Say what is on your mind and an organised Markdown note appears in your Obsidian vault.
beta-idea.md
Meeting assistantin testing
It records your microphone and system audio separately, so on headphones it tells your words apart from everyone else's. The conversation becomes a note with a summary, a list of decisions and the full transcript.
Me: I'll send the test buildGuest: I'll confirm by Friday

What we stand on

Runs locally, offline too

The transcription and summarisation models run on your own machine. Recordings and notes are never sent anywhere, so we have no access to them. They never reach us in the first place.

Built for Polish

We choose and test models on Polish recordings. What counts for us is how they handle inflection, proper names and casual speech, not a claim of support for dozens of languages.

Plain Markdown files

Notes are saved as .md files in your vault. You can open them in any editor, including after you stop using our app.

The app is in beta on macOS, with Windows on the way. Download it below — and if anything breaks, tell us.

Download

macOS beta

The app itself is 11 MB. The transcription and summarisation models, 6.5 GB together, are downloaded on first launch and the app walks you through it. After that everything works without an internet connection.

Download the beta

version 0.9.4 · 11 MB · .dmg

The package is signed with an Apple Developer ID certificate and notarized, so macOS opens it without warnings. Drag the app into Applications after downloading.

What you need

  • macOS 14 (Sonoma) or newer
  • Apple Silicon — M1 or newer
  • 16 GB RAM for the full version
  • 7 GB of space for the models (1.5 GB in transcription-only mode)

Less than 16 GB? On first launch pick the “Transcription only” profile — it downloads 1.5 GB instead of 6.5 GB and never starts the summarisation model, the only part that needs that much memory. Dictation and voice notes work as usual; meeting summaries and knowledge curation are gone. We have not measured this on an 8 GB machine yet — if you try it, tell us how it went.

The beta is free for the whole testing period. Version 1.0 will be paid — leave your address when downloading and you will hear about the release first, with a beta tester discount. If something does not work, write to us through the form below. We read every report.

For teams

When the recording cannot leave your office

In some professions, sending a recording to an outside service is simply off the table. VantoLabs has no technical means of doing it. The app connects to no provider, stores no API keys and requires no account. The recording, the transcript and the note stay on your disk, including when you work without an internet connection.

Therapy and psychology
You dictate the note after a session and the app organises it on your own machine. The recording never reaches an outside provider, so there is no data processing agreement to sign.
Law firms
Case discussions, client conversations and dictated drafts. The content never leaves the machine, and notes are saved as plain files you store under your own rules.
HR and recruiting
Turn candidate conversations into a summary and a list of decisions, without sending the recording to an outside tool.
Consulting and accounting
Save what you agreed with a client as a note with a task list. Financial data does not pass through an external server.

Rolling it out with you

We can prepare a glossary of the terms used in your work, train the team and fit the note format to your document flow. Write to us through the form below.

Let's talk about a rollout

What we describe here is how the app behaves: data does not leave your device. Whether that satisfies your professional obligations and GDPR duties is your call. We are glad to answer any technical question your counsel may have.

Direction

Where this is headed: a second brain

in the making

Today the model organises a single note. We are working on a layer that reads the whole collection and builds knowledge pages from it, covering topics, people and projects, with links back to the source notes. Your recordings and transcripts stay untouched, because the curator writes its results to a separate folder.

The second thing we are building is asking questions of your own notes. A keyboard shortcut opens a small window, you type a question about decisions, people or plans, and the answer comes back with links to the notes it came from.

_inbox — your notes
LLM curator
_wiki — knowledge pages

Shipping in the beta

  • Dictation in any application
  • Voice notes straight into Obsidian
  • Meetings split by who was speaking
  • Curator: notes → knowledge pages

What we are building

  • Ask your own knowledge — a Spotlight-like window
  • Reliable speaker attribution on speakers too
  • A Windows version

Founders

Who we are

VantoLabs is two founders and one product that we use ourselves every day. We set the direction and the plans together. What differs between us is skills, not standing. We have ideas for further products, but we want to finish this one first.

PP

Paweł Pelar

Co-founder. AI and data engineer — he builds production GenAI systems (LLMs, speech analytics, RAG) and cloud data platforms, from infrastructure to deployment. Holds an M.Sc. in Artificial Intelligence and Azure/Databricks certifications. At VantoLabs he is responsible for architecture and product engineering.

TK

Tomasz Korzeniewski

Co-founder. A business and operations practitioner — he makes processes actually work, and AI is his everyday working tool, not a curiosity. At VantoLabs he is responsible for priorities, testing and operations — making sure the app solves a real problem, not just runs.

You can also find us on GitHub

Get in touch

Contact

Have a question, hit a bug, or thinking about a rollout at your firm? Write to us and we will answer personally.

By sending the form you agree to the processing of your e-mail address for the purpose of replying. Details in the privacy policy.