Use cases
Dictation your security team can actually sign off
Most dictation tools fail a security review for one reason: the audio goes to a vendor. This page is what to send IT when it does not, including what we do not claim.
Last updated
The question a review actually asks
Security reviews of dictation software converge on one thing: where does the audio go, and who else can hear it? Everything else — encryption, retention, subprocessors, breach notification — exists because the answer is normally "to the vendor".
When recognition runs on the endpoint, most of that questionnaire stops applying. Not because we have good answers, but because the questions are about a transfer that does not happen.
| Review question | Cloud dictation | VV |
|---|---|---|
| Is audio transmitted to the vendor? | Yes | No |
| Where is it processed? | Vendor infrastructure | The user's Mac |
| Is it retained? | Per vendor policy and settings | Locally only, for a period the user sets |
| Is it used to train models? | Per vendor policy and settings | No — we never receive it |
| Subprocessors handling voice data | Typically several | None |
| What does a vendor breach expose? | Whatever is held | No dictation content — we hold none |
| Does it work with no connection? | No | Yes |
What we do not have
Listing this properly is more useful than a sales conversation that discovers it in week three.
- No certifications. Not ISO 27001, not SOC 2, not HIPAA-attested. We are a small UK company and we claim none of them. If a certification is a hard gate, we will not pass it.
- No SSO or SCIM. Licences are per user, managed by the individual or by whoever bought them.
- No MDM configuration profile. Settings are per machine. You cannot centrally enforce, for example, a maximum history retention.
- No admin console. No central view of who is using what.
- No audit log. Dictation history is local to each Mac and visible to that user.
- Mac only. Apple Silicon (M1 or later), macOS 14 Sonoma or later. Windows is in development with no date promised.
What deployment looks like
- One machine first Install, complete first run, verify the architecture by disconnecting the network and dictating. Record the date.
- Check your accessibility permission policy Dictation needs accessibility permission to type into other applications. Managed Macs often restrict this, and it must be allowed or nothing works. This is the most common blocker.
- Decide the retention position Users can set history retention themselves. Decide what you want it to be and tell them, because you cannot enforce it centrally.
- Confirm FileVault is enforced The local history sits on the disk. FileVault is what protects it, and you probably already require it.
- Decide on cloud rewriting An optional rewrite step can be pointed at a cloud provider. If that is not acceptable, say so in the rollout note — the default is local.
- Pilot with five people for a fortnight Then decide. Dictation adoption is highly individual; some people take to it immediately and some never do.
Facts for a security questionnaire
- Recognition runs on the endpoint; no dictation audio or text is transmitted to the vendor.
- The speech model is downloaded once from a pinned source and verified by checksum before use.
- The application is signed with a Developer ID certificate issued to Vaitly Limited and notarised by Apple.
- Network activity is limited to model download, licence validation, update checks and optional product analytics, which can be turned off.
- Dictation history and audio are stored locally, under the user's Library, for a user-configured retention period.
- An optional rewrite step runs locally by default; a cloud provider can be configured by the user and is labelled in the interface.
- Vaitly Limited is a UK company, number 16814185.
- No security certifications are held or claimed.
Who this suits
- Professional services firms. Law, accountancy, consultancy — client-confidential material and a partner who has to sign it off.
- Healthcare administration. With the caveat that we hold no HIPAA attestation; the architecture helps, the paperwork does not exist.
- Security and engineering teams. Where the content of a note is the thing being protected.
- Organisations with air-gapped or restricted machines. On-device recognition works where nothing else does.
- Small teams without a procurement department. Where architecture is the assurance and a certification would be theatre.
Who it does not suit: organisations needing central policy control, a certification, SSO, or Windows today. See dictation for confidential work for the decision framework.
Licensing
Pro covers 2 active devices per licence, which handles the laptop-and-desktop case. For a team, buy volume rather than individual seats — talk to us rather than putting twenty purchases through expenses.
Lifetime options exist (£29.99 for Founder Lifetime, £79.99 for Pro Lifetime), which some organisations prefer to a renewal they have to justify annually. Prices are in pounds sterling; VAT is added at checkout where it applies. Every purchase is refundable in full within 14 days.
Questions
Is VV ISO 27001 or SOC 2 certified?
No. We hold no certifications and claim none. If your process requires one, we will not pass, and we would rather tell you now.
Do you sign a data processing agreement?
For dictation content there is nothing to process — the audio and text never reach us. We do hold account and billing data for purchased licences, which our privacy policy covers and which we will discuss.
Can we enforce settings centrally?
No. There is no MDM configuration profile and no admin console. Settings including history retention are per machine and per user. If central enforcement is required, we are not the right fit.
Does it need accessibility permission on managed Macs?
Yes, and this is the most common deployment blocker. Typing into other applications is exactly what that permission governs on macOS, and there is no narrower alternative. It must be permitted by your management policy.
What happens if the vendor disappears?
Installed copies keep working, because the model and the data are already on each Mac. Nothing would be fixed, but nothing stops. That is a genuine difference from a cloud service, where the product ends when the servers do.