Local-first transcription tools: nine compared

"Runs on your own machine" used to mean two or three obscure applications. It now means about nine, and they do not cost the same, do not label speakers equally well, and do not suit the same people.

This is a buyer's comparison written by the people who make one of them. We name the three cases where somebody else is the better purchase, because you would find out anyway.

Everything here comes from our own research notes, dated July 2026. Where our notes record nothing, the table says "not stated" rather than guessing.

The table

Tool	Price and model	Speaker labels	Open source	Platforms
Meetily	MIT core, Pro $120/user/yr (annual, 1 device)	"Soon" per its pricing page, Jul 2026	Yes (MIT)	Desktop only
OpenWhispr	Free	Yes, local (pyannote + CAM++)	Yes	Desktop only, cross-platform
Talat	One-time purchase, amount not stated	Yes, on-device	Not stated	Mac only
Scriberr	Free, self-hosted	Yes	Yes	Self-hosted server
Anarlog	Free	Via the models it uses	Yes	Not stated
MacWhisper Pro	€59 one-time, free tier below it	Yes	Not stated	Not stated
Superwhisper Pro	$249.99 lifetime, or $8.49/mo, $84.99/yr	Not stated	Not stated	Not stated
Handy	Free	Not stated (dictation utility)	Yes, MIT-ish	Not stated
yapyap	$69 one-time, 2 devices, free trial	Yes, channel-aware	No	Desktop plus iOS and Android companion

What actually leaves your computer

Here is the finding that surprised us. On the question everybody asks first, every tool in that table gives the same answer. Your audio stays on your machine. Privacy is the price of entry in this category now, not a reason to pick one product over another.

So the question is not "does it stay local". It is what else the tool needs from you: an account, a network connection at some point, a subscription that keeps running, a company that has to stay in business.

Three things do leave your computer across these tools, and they are worth knowing about. Speech models get downloaded the first time. Applications check for updates. And if you deliberately connect a hosted language model for the summary step, the transcript goes wherever you pointed it. That last one is a choice you make, not a default.

Pick the one that matches your job

Buy MacWhisper if you are on a Mac and want the least money. €59 one-time against our $69, with local transcription across all model sizes, speaker labels, meeting capture, batch processing, subtitles and a free tier underneath. If that is the whole of what you need, buy it. We have no argument against it.

Try Meetily if you want to read the source. It is the closest thing to a direct rival: same architecture, real traction as a self-hosted notetaker, MIT core you can fork. Check its speaker labelling before you commit, because its pricing page still listed that as "soon" in July 2026. Pro is $120 per user per year and covers one device.

Try OpenWhispr if your budget is zero. Free, cross-platform, local speaker labels, custom prompts, chat across your notes and calendar sync. The most feature-dense free option here.

Try Scriberr if you already run a home server. Transcription becomes one more service on a machine you already maintain.

Talat is Mac only, sold once, and runs on Apple's neural chip. Anarlog leans on file ownership and your own provider keys. Handy and Superwhisper are dictation tools rather than recorders, so judge them on how well they type for you and not on this list's terms.

Where yapyap loses

Our own launch review records the gaps. Publishing a lens is a manual pull request rather than one tap. There is no statistics panel for talk time or words per minute. There is no Obsidian or Logseq export, though Markdown export exists. Desktop retention and auto-delete are not built. Dictation works inside the ⌘K box but not system-wide. And there is no free tier, only a trial, which is the weakest part of our pricing against three rivals that do have one.

Where yapyap wins

One recording becomes many write-ups, through lenses you can write and install from other people. Your phone records on its own away from your desk and hands the file to your desktop over your own network, which no local-first rival in our research offers. And speaker labelling keeps your microphone and the call audio on separate tracks instead of guessing from one mixed track.

Sources

• whisper.cpp — the local speech engine many of these tools use.
• Meetily — open local meeting notes (check its site for current pricing).
• Vendor pricing changes. Treat these figures as notes dated in the post, not live quotes.