What cloud dictation actually sends, why offline speech to text on Windows is now practical, what Saytify does and does not transmit, the PDPL questions to ask any dictation vendor, and a two-minute test you can run yourself.
Dictation is intimate in a way typing is not. When you type an email, the only record is the email. When you dictate it with a cloud service, there is also a recording of your voice, in your room, saying your client’s name, your price, and the thing you decided not to put in writing after all. Where that recording goes, how long it stays, and who can listen to it are not footnotes. They are the product.
This post is about why Saytify processes speech on your own PC, what that means in practice, and how to evaluate any dictation tool, ours included, before you trust it with your work.
What cloud dictation sends
A typical cloud dictation app captures audio from your microphone and uploads it to the vendor’s servers, where speech recognition and text clean-up run. The text comes back and is typed into your app. From your side it feels local. It is not.
That means, at minimum, the vendor receives your raw voice recording and the resulting text. Depending on the service, it may also receive the name of the app you were typing into, a screenshot or window title for “context”, your contacts or dictionary for better recognition, and usage analytics. Vendors vary enormously in how long they keep this, whether humans may review it, and whether it trains future models. The only way to know is to read the policy, and the only way to be sure is for the data never to leave.
For a lot of work, that is a deal-breaker. Tender pricing, legal correspondence, HR decisions, patient notes, unreleased financials, government projects, anything under an NDA: none of it should pass through a third party just so you can avoid typing.
Why offline is now practical
For years, offline speech recognition meant worse accuracy and a slow, awkward experience. Two things changed.
First, open speech models became good enough to run locally and still handle real-world audio in multiple languages. Second, ordinary Windows PCs got fast enough, through consumer GPUs and better CPU inference, to run those models in near real time.
Saytify is built on that shift. Speech recognition, the second AI pass that cleans up the text, and the formatting all run on your machine. On my own PC, text appears about two seconds after I release the talk key (median over real dictations), with recognition streaming while I speak. On a PC with less memory, Saytify picks lighter models automatically. The trade-off that used to exist, privacy versus usability, has mostly gone.
What Saytify does and does not transmit
I want to be precise here, because “offline” is a word that gets stretched.
Never transmitted, under any setting:
- your audio, while you speak or afterwards;
- the recognised or corrected text;
- your dictation history, personal dictionary or settings;
- the names of the apps you dictate into, screenshots, or window titles;
- usage analytics or telemetry. Saytify contains none.
Transmitted, and only this:
- At activation, once per device: your license key, a one-way hash derived from your Windows installation and system drive (so we can count devices without knowing your hardware), the device name, Windows version and Saytify version.
- About every 30 days, a silent license check with the key and that same device hash. Between checks the app works fully offline. If your PC stays offline past the current period, the app asks you to connect briefly to renew.
- During setup, the speech models are downloaded once from our servers.
That is the complete list. The Privacy Policy spells it out formally, including the data we hold about your purchase (email, order and payment reference; card details are handled only by Moyasar).
Why we still check the license
A fair question: if the app is offline, why does it talk to us at all?
Because a Lifetime license is sold once and used for years, and we need a way to honour refunds and chargebacks, keep one key from being shared across a whole office, and let you move your license to a new PC. A periodic check, carrying nothing but the key and the device hash, is the least invasive way to do that. It is also the only network connection the app makes.
Test it yourself in two minutes
Do not take a vendor’s word for it, including ours. Any offline claim can be checked:
- Install and activate the app while connected.
- Disconnect from the internet: turn off Wi-Fi, unplug the cable, or enable airplane mode.
- Open Notepad and dictate a few sentences in Arabic and English.
If the text appears, speech recognition is running on your machine. A cloud service will fail or hang at step 3. For a stricter test, block the app in Windows Firewall for a month and watch it keep working until the next license check asks for a connection.
Questions to ask any dictation vendor
Whether you are choosing for yourself or writing a procurement checklist, these are the questions that matter, and the answers Saytify gives.
- Where is my audio processed? On your own PC. It never leaves.
- Do you store recordings or transcripts? No. We never receive them.
- Is my data used to train models? No, and it cannot be: we do not have it.
- What leaves the device, exactly, and how often? A license check with the key and a device hash, about every 30 days. Nothing else.
- Who are your sub-processors? Moyasar for payments, Cloudflare for hosting and security, our email provider for transactional email. They see order and license records, never your voice or text.
- Can I use it with the internet off? Yes, after the one-time setup.
- Can I delete my data? Your dictation history is on your PC; clear it in the app. Your purchase record can be deleted on request, subject to the tax records we must keep; email support@zworks.net.
- Where is the company and which law applies? Zworks, Jeddah, Saudi Arabia. The Saudi Personal Data Protection Law (PDPL) applies to how we handle your account data, and our privacy notice is published in Arabic and English.
A note on Saudi PDPL and regulated work
The PDPL has been enforced since 2024, and proposed amendments under consultation this autumn point toward stricter rules on storing personal data inside the Kingdom and on cross-border transfers. If you handle personal data of people in Saudi Arabia, a cloud dictation service that ships their voices abroad is a question you will eventually have to answer to your own compliance team.
An offline tool sidesteps most of that question. Your dictated text is created on a device you already govern, under the same controls as the documents you type. There is no new processor, no new transfer, and no new vendor assessment just for dictation. This is not legal advice, and your compliance officer should confirm it for your context, but it is a much shorter conversation.
How the alternatives stand, as of October 2026
Comparisons go stale, so I will date them. As of October 2026: Windows’ built-in voice typing (Win+H) uses online speech recognition, needs an internet connection, and does not list Arabic among its supported languages. Wispr Flow processes audio in the cloud and has no offline mode. Superwhisper runs locally, offers a lifetime license at $249.99, and does not name Arabic as a supported language. Saytify runs offline, handles Arabic and English in the same sentence, and is sold as a one-time Lifetime license with a Founders price of SAR 499 for the first 300 buyers.
Privacy should not be the premium tier
The industry pattern is to offer “on-device processing” as an enterprise feature or an upsell. I think that is backwards. Speech recognition is fast enough to run on a normal PC, so that is where it should run, for everyone, at the base price. That is how Saytify is built, and it is not going to change.
If you want to see the rest of what it does, the features page is the place to start; engineers who live in AutoCAD should read the engineers page. And if you have a privacy question I have not answered here, email support@zworks.net. I would rather answer it before you buy than after.
