HoldSay: A Wispr Flow Alternative with On-Device Dictation and No Subscription
HoldSay is an alternative to Wispr Flow for people who want to dictate into any app, but would rather run speech recognition on their own machine and pay once instead of paying a monthly subscription. It also handles two jobs Wispr Flow does not: recording calls or meetings with notes, and transcribing existing audio or video files. I built it with my co-founder, Legolas Wang, because those were the tools we kept paying for ourselves.
One floating card for dictation and meeting notes, no subscription.
Wispr Flow is good at its central job. It is polished, widely liked, and makes desktop dictation feel natural: speak, then get clean text where your cursor already is. Its documentation is also upfront about its pricing and cloud architecture. This is not a page arguing that Wispr Flow is secretly something other than what it says it is.
It is a page for the slightly particular person who likes dictation already, has reached the 2,000-word weekly limit on the free desktop tier or is looking at another monthly charge, and would prefer their spoken words to stay on their computer.
The practical difference between Wispr Flow and HoldSay
With Wispr Flow, you pay for a cloud dictation service. Flow Pro is listed at “$15 per user per month” or “$12 per user per month billed annually”. At list price, that comes to $180 per year, or $144 per year when billed annually.
With HoldSay, you buy one device seat. The regular price is $69.99. We also have a limited early-bird price of $39.99. There is a free two-day trial, with every feature turned on and no card required.
The bigger difference is where speech recognition happens. Wispr Flow’s security materials say it “runs entirely in the cloud”. Audio is streamed to the backend and is not persisted locally. Under its default settings, its audio and transcription data “may be used to evaluate, train, and improve” Wispr’s models.
HoldSay takes the opposite technical route. The HoldSay Dictation model runs on your own machine. Your audio, recordings, and transcripts stay on your machine and are never uploaded.
That choice has consequences. A cloud service can feel pleasantly invisible because there is no model to download and no local library to consider. HoldSay needs a one-time download on first launch; the HoldSay Dictation model takes about 2 GB on disk. In return, you can dictate after your Wi-Fi has disappeared, which is more common than software websites tend to acknowledge.
| Tool | What it does | Cost | Where the audio goes |
|---|---|---|---|
 Wispr Flow |
|
Free tier: 2,000 words per week on desktop. Flow Pro: “$15 per user per month” or “$12 per user per month billed annually” | Audio is streamed to the backend, and under default settings it may be used to train their models |
 HoldSay |
|
One payment per device seat: $69.99 regular price; $39.99 limited early-bird price | Nowhere. Audio, recordings and transcripts stay on your Mac or PC. No account, no training on your data |
Why we made HoldSay
Legolas and I have been indie developers for six years. Earlier this year, we made a small and mildly annoying list of the subscriptions we used to talk to our computers.
One was for dictation. Another was for meeting notes. We were paying for Granola’s Business plan at $14 per user per month, and its notes were stored on AWS in the US. It trains on anonymized data unless you opt out. Neither product was doing anything unreasonable. We just kept wondering why these were separate cloud services when the useful parts could live in one small card on our screens.
So we built that card.
Hold a key, speak, then release it. The text goes into the app in front of you, at the cursor. I use it in email, issue trackers, code comments, documents, and the small text field where I am supposed to explain why a charge on my card is legitimate. HoldSay does not need a special editor.
Dictating a prompt straight into a chat box.
The card also records. For lectures, interviews, and calls, HoldSay can capture system audio so both sides of a call are included. No bot joins the meeting. When the recording is done, it can create a transcript, apply speaker labels from voice profiles that never leave the machine, produce an on-device summary, and run a Polish pass that removes filler words without changing what you meant.
There is a useful detail for long recordings. LiveScribe transcribes while recording is still underway. A two-hour interview is already transcribed when you press stop, rather than becoming a two-hour job for later.
Finally, HoldSay transcribes files. Drag an audio or video file onto the card, or right-click it in Finder. If you have a folder of files, you can send the whole folder at once. I have found this especially handy for recordings I had already accumulated in a directory with names such as final-final-actually-final.m4a.
Everything ends up in ~/HoldSay Library by default. Each note gets its own dated folder containing the audio, transcript, and meta.json. There is no account, login, or database hiding behind it.
Dictation is the part I use most
The recording and file tools matter, but ordinary dictation is what I open HoldSay for on a normal day. After about two months, my own stats page shows 840,904 words dictated and recorded across 5,703 dictations and 281 recordings. On a typical day, I dictate more than a hundred times. Some of those recordings run for hours.
My dictation history: past 875,000 words by the time of this screenshot.
HoldSay supports dictation in 30 languages, all on-device and without a higher tier for particular languages. You can also import names and jargon into the custom dictionary from a CSV. That matters when your work contains product names, client names, technical terms, or surnames that the rest of the world has elected to spell differently from your contact list.
How fast is HoldSay dictation?
Across 221 consecutive dictations on one MacBook Air M3 with 24 GB of memory, running macOS 26.2, on August 3–4, 2026, the median delay from key release to text on screen was 3.1 seconds. The 90th percentile was 4.4 seconds, and 92% completed within five seconds. That is one machine over two days, not a promise about every computer in every room.
For long dictations, HoldSay starts transcription in the background if you hold the key for about twenty seconds. Keep speaking for two minutes, then release the key, and much of the work has already happened.
Dictation and recording share one resident model, and a Mac with 8 GB of memory stays comfortable. The installer itself is about 107 MB because the model downloads when first used. Minor updates are small deltas, between 0.7 MB and 1.6 MB.
Does HoldSay work offline?
Yes. Install HoldSay, let the HoldSay Dictation model download, turn on Airplane Mode, and dictate. It still types what you say.
This is the first test we run on our own builds. It is a plain test, and it works with any dictation app: remove the network, then see what still happens.
HoldSay touches the network exactly three times: for the one-time model download on first launch, license verification, and update checks. None of those carries audio, transcripts, or personal data. After the model has downloaded, dictation, recording, transcription, summaries, speaker labels, and the Polish pass run on the machine.
Offline does not mean “your files magically appear elsewhere.” Nothing syncs. Your notes live on one machine unless you move the folder yourself. That is intentional, but it is also a limitation worth taking seriously before buying.
Who should stay with Wispr Flow
Some people should stay with Wispr Flow.
Stay with Wispr Flow if you use an Intel Mac. On the Mac side, HoldSay requires Apple Silicon and macOS 14 or later; Intel Macs are not supported. HoldSay also runs on Windows PCs, though this article comes from my Mac experience.
Stay with Wispr Flow if its free 2,000 words per week already cover your needs. There is no reason to buy software to solve a limit that is not actually bothering you.
And stay with Wispr Flow if you prefer a cloud service that you never have to think about in terms of local storage. HoldSay gives you visible files in a visible folder. That is the point for some people, and a chore for others.
Teams that need a shared library which syncs itself should also not buy HoldSay. It does not provide one. You can move the files yourself, but that is different from collaborative cloud storage, and pretending otherwise would waste your afternoon.
Is buying once better than a subscription?
I do not think subscriptions are inherently bad. They can make sense for services with ongoing server costs, shared data, or features that need a central system. But dictation on one computer does not have to be that arrangement. For us, the cleaner deal was a local tool with a one-time price.
A HoldSay seat activates one Mac or one Windows PC. Multi-seat licenses can mix both platforms. The price is per device seat, not per person wandering across a pile of laptops.
That is HoldSay’s offer: a floating card for dictation, recording, notes, and transcription, with the work staying on your machine.
FAQ
Is HoldSay a replacement for Wispr Flow?
For on-device dictation into any app, yes. HoldSay also records with meeting notes and transcribes files, while Wispr Flow does not offer those features.
How much does HoldSay cost?
$69.99 per device seat, paid once; $39.99 is a limited early-bird price. There is no subscription, and the free two-day trial has every feature on with no card required.
Is my audio uploaded or used to train models?
No. The HoldSay Dictation model runs on your machine, and audio, recordings and transcripts never leave it. The only network traffic is the one-time model download, license verification and update checks, none of which carries audio or transcripts.
Does HoldSay work on Windows?
Yes. Each seat activates one Mac or one Windows PC, and multi-seat licenses can mix both. The app does the same four jobs on either platform; this article was written from the Mac side.
Does it work on an Intel Mac?
No. On the Mac side HoldSay needs Apple Silicon and macOS 14 or later; an Intel Mac cannot run it, though a Windows PC can.
Which languages does dictation support?
Thirty, all running on-device, none reserved for a higher tier. The full list is in the FAQ, and a custom dictionary imports your names and jargon from a CSV.
What happens to my notes if I stop using HoldSay?
Nothing. They were never in a database to begin with: each note is a dated folder holding the audio, the transcript and a meta.json, readable by anything, forever.
Was this page accurate? If something here doesn’t match what you see on your Mac, tell us at support@holdsay.com and we’ll fix the page.