HC Healthcare evaluation
On-device dictation for clinical teams.The transcription vendor leaves the data flow.
Recognition and cleanup run on the Mac, so no transcription provider receives clinical speech. Compliance is decided by your deployment. Here is what the software does, and how to evaluate it with synthetic data.
Audio never saved or sentNo transcription vendorAES-256-GCM at restCopy only mode
01 Processing location
One less party in the path.
When dictation is transcribed in a vendor cloud, that vendor's infrastructure, retention, subprocessors and contracts enter your risk analysis. On the Mac, they don't.
02 The product
In clinical terms.
What happens to a dictation on a managed Mac, from the microphone to the field it lands in.
| Property | Behavior |
|---|---|
| Requirements | Apple silicon Macs, macOS 15 or later, about 484 MB for the speech model. |
| Recognition | Parakeet TDT v3 through Core ML, on the Mac. |
| Transcription vendor | None. The app does not send audio or transcripts to LockedIn Labs or anyone else; after the paste, text follows the receiving app's policies. |
| Network | Never downloads models, checks for updates, sends telemetry or calls a hosted service. Its speech library's downloader is switched off before any model loads. |
| Audio | Held in memory, never written to disk. |
| History | In memory until you quit, by default. Seven retention options, from “Do not retain” to “Keep indefinitely”; AES-256-GCM on disk. |
| Delivery | Saved to History first, then one paste at the cursor; the previous clipboard is put back. |
| Password fields | Fields macOS identifies as password fields are skipped: nothing typed, nothing copied. |
| Copy only | Nothing is typed into other apps until you choose Copy. Works without Accessibility. |
| Cleanup | Local rules, written for English. The optional Apple Intelligence pass (macOS 26 with Apple Intelligence) is off by default and falls back to the rules when it fails a meaning check. |
| Vocabulary | Your terms as reviewable, removable rules, stored encrypted. Learning from edits is off by default. |
| Accounts | None. No sign-in, no analytics, no telemetry. MIT-licensed, no activation. |
03 Evaluation
Run the pilot on synthetic data.
Start on managed Macs with a team build and synthetic dictation. Plan the pilot around six points, owned by your clinical, security and IT leads.
- 01
Synthetic first.
Dictate made-up content into the apps your teams use. No patient information until your organization approves production use.
- 02
Approved terminology.
Build vocabulary from an approved term list. Keep it as one CSV you can review, import, and hold under your own change control.
- 03
Clinician review.
Keep review of high-risk content in the workflow: numbers, negation, laterality, medications and doses, before the text is relied on.
- 04
Verified delivery.
Test delivery into your record-system and browser fields: focus changes, password fields, and re-inserting from History.
- 05
Defined retention.
Choose a History retention that fits policy, decide whether meeting notes are used at all, govern clipboard tools, and require FileVault and screen lock.
- 06
Acceptance thresholds.
Agree accuracy and safety thresholds in advance, measure them on representative workflows, and review them before production use.
04 Beyond healthcare
Same properties, other regulated work.
Legal, finance and public-sector teams that keep dictation off the cloud evaluate it the same way: no transcription vendor, no internet once the speech model is installed, explicit vocabulary, and encrypted local retention with an off switch.
05 Questions
What clinical teams ask first.
How does LockedIn Flow relate to HIPAA?
HIPAA compliance is a property of a deployment: your organization's risk analysis, policies, device controls, workforce practices, contracts, incident response and the systems text lands in. LockedIn Flow's part is architectural. Dictation audio and transcripts are processed on the Mac, so no transcription vendor receives them.
Does clinical dictation go to a vendor cloud?
No. Dictation audio and transcripts are processed on the Mac and are not sent to a transcription provider. The app never downloads models, checks for updates, sends telemetry or calls a hosted service; its speech library's downloader is switched off before any model loads.
After the paste, text follows the policies of the app that received it. The security architecture covers the full data flow.
How should a clinical team start an evaluation?
Request a build for managed Macs, dictate synthetic text into the applications the team uses, and record the results against a pilot plan. Keep clinician review of dictated text in the workflow.
HC Healthcare evaluation
Plan the evaluation with us.
Clinical and security teams can share a non-sensitive outline. Never send patient information, recordings, private terminology, credentials or production configuration through the website.
Architecture statements reviewed against the source · October 2026