AI Clinical Documentation That Runs Locally: A Privacy Guide

Clinical documentation is the written record of a patient encounter: the history, the findings, the assessment, the plan, and any follow-up. It is the backbone of care, billing, and continuity, and it is also the part of the job that eats the most unpaid time. Surveys of physicians have found they spend roughly one to two hours on documentation for every hour of direct patient care, and a large share of that spills into evenings.

AI clinical documentation software exists to shrink that gap. In practice these tools fall into a few groups. Some are ambient scribes that listen to the whole visit and draft a note from the conversation. Some are structured template tools inside an electronic health record. And some, the category this guide sits in, are voice-to-text tools that turn a note you deliberately speak into clean, formatted text you can file.

The common thread is speech. You talk, the software listens, and an AI layer produces readable text instead of a raw transcript full of filler. The difference that matters for privacy is not what the software produces. It is where the listening happens.

Yaps is a voice-to-text tool in that third group, built privacy-first. You trigger it with the Yaps hotkey, speak your note, and it does three things on your own device: it recognises the speech, it cleans up the result, and it saves the note into a private searchable store. No part of that requires sending your voice anywhere. For a deeper look at the dictation engine itself, see the dictation feature overview, the voice tools for healthcare page, and the HIPAA dictation workflow use case.

Most documentation tools process speech in the cloud because it is easier to build that way. A phone or a laptop sends the audio to a data centre, a large model transcribes and rewrites it, and the text comes back. It works well and it is why so many tools sound impressive out of the box.

For patient notes, that convenience carries a cost that is easy to miss. A dictated clinical note is not ordinary text. It contains names, ages, diagnoses, medications, dosages, and treatment plans, which is the most regulated category of personal information there is. The moment that audio is transmitted, a copy of it exists on infrastructure you cannot see, audit, or control.

From there, the familiar risks follow. The vendor may retain the recording, or use it to improve its models unless you read the terms and opt out. The vendor can be breached, and healthcare has for years carried the highest average data breach cost of any industry. A vendor can also receive a legal request for data you assumed was private. None of these require bad faith. They are simply what becomes possible once a copy of the note leaves your control.

There is also a contractual layer clinicians discover late. A cloud tool that processes patient information for a covered entity generally needs a signed agreement governing that data, and a consumer privacy policy is not the same thing. That is work a solo clinician or small practice rarely has the appetite for.

Here is the shift in one sentence. If the audio and the notes never leave the device, no vendor breach can expose what is not there.

That is a different kind of protection from the usual one. Most privacy features add safeguards on top of a system that still moves your data around: encryption in transit, access controls, retention limits, audit logs. Those matter, but they are all promises about how carefully a third party will hold a copy of your patient's note. On-device capture removes the copy. There is no transmission to secure, no server-side store to lock down, and no external party whose diligence you are trusting.

This changes what you have to reason about. Instead of evaluating a vendor's security posture, its subcontractors, and the fine print about model training, the surface you are protecting is much smaller: your own device, which you already secure with a password, disk encryption, and a lock screen. The custody chain for a single-clinician workstation is about as simple as it gets.

You cannot leak a recording you never uploaded. On-device capture protects patient notes by removing the copy, not by guarding it more carefully.

Yaps

None of this makes your obligations vanish. Keeping notes on device means the device itself has to be secured properly, and this guide is not legal or compliance advice. But it moves the hardest part of the problem, the part where patient audio sits on someone else's server under terms you did not write, off the table entirely. For the regulatory background in plain language, our HIPAA-compliant dictation guide and the broader voice privacy in regulated industries piece are the best next reads, and the healthcare dictation privacy article covers where patient voice data actually goes.

The point of an on-device tool is that the privacy is invisible in daily use. You are not managing it. You are just writing notes faster. Here is what that looks like with Yaps across a clinic day.

You dictate the note. After an encounter, press the Yaps hotkey and speak the note the way you would say it to a colleague. On desktop the trigger is the Fn key, and on Android it is the dictation button on the Yaps keyboard, so you can capture a note on a phone between rooms. Recognition runs on the device, so there is no lag waiting on a network and no failure when the clinic Wi-Fi is patchy. It also works with the internet switched off, a clean way to verify nothing is uploaded: if it still works offline, the processing is genuinely local.

Dictation is multilingual. Yaps recognises around 25 languages and detects the language from your speech, so a clinician who documents in more than one language does not have to flip a setting mid-note. This matters in real clinical settings where a patient's history might come out in one language and the plan in another.

On-device cleanup fixes the ramble. Spoken notes are messy. You restart sentences, you say "um," you correct yourself. Yaps cleans that up locally and on by default: it removes filler words and self-corrections, fixes punctuation and capitalisation, and formats spoken lists and numbers into actual lists. What was a spoken jumble reads like something you sat down and typed, without a round-trip to a server to make it presentable.

The finished text goes where you need it. The cleaned note can drop straight into the field you are typing in, or it can be saved into your private vault. To be precise about what Yaps is and is not: it types into whatever field has focus, the same way any keyboard does, but it does not connect to, read from, or write structured data into your EHR. It is not an EHR plugin. You move the finished note into the record yourself, which keeps the tool simple and keeps Yaps out of the systems it has no business being inside.

You can also have a note read back to you. Yaps includes read-aloud, with 18 voices on desktop and 2 on mobile, so hearing a note before you file it can catch an omission your eye slides past. One honest caveat: the shipped voices are English speakers in practice, so if you need notes read aloud across many languages, a dedicated multilingual cloud reader is the better fit there. For dictation, the multilingual support stands.

The whole workflow above reduces to three moves you will do without thinking after the first day.

1. Dictate with the Yaps hotkey. Press the hotkey after the encounter and speak the note the way you would say it aloud. Recognition happens on the device, offline if you want, so the audio is never transmitted.
2. Let on-device cleanup format it. Cleanup runs locally and on by default, stripping filler and false starts, fixing punctuation and capitalisation, and turning a spoken list into a real one.
3. Save it privately. Send the finished text into the field you are working in, or store it in your local vault. There is no upload, so the note stays on your machine.

Three steps, no server in the loop, and no account required for the core capture. You only need an account for billing.

It helps to see the three main options side by side, because they are genuinely different tools. Yaps leads here because privacy is its whole design, but the table is honest about where the others are the stronger pick.

Read the table as a fork, not a scoreboard. The rows where Yaps says "no" are not gaps to apologise for. A tool that does not upload audio cannot leak it, and one that does not draft a note from the live room does not have to record it. The next section covers the cases where those "no" rows genuinely point you to a different tool.

Being useful means being clear about the edges, so here is the honest boundary. Yaps captures, cleans, and organises notes you speak, and it transcribes audio files you already have. It does not do everything the phrase "AI clinical documentation" can imply.

Yaps does not join or transcribe a live visit. There is no bot that sits in the room or in a telehealth call, listens to the whole conversation, and drafts a note from the dialogue. That ambient-scribe capability is a real and growing category, and it is a different tool from a private voice-to-text app. If your practice needs an ambient scribe that captures the visit as it happens, Yaps is not that.

Yaps does not integrate with your EHR. It types cleaned text into whatever field has focus, but it does not connect to your record system, does not pull the patient chart, and does not write structured fields or codes back into it. If you need documentation that flows into specific EHR fields automatically, that is an EHR-integrated product, not this.

Yaps does not do automated medical coding or clinical decision support. It will not suggest a billing code, flag a drug interaction, or offer a diagnosis. It writes down what you say. It is not a medical device and nothing it produces is medical advice.

So the honest way to place Yaps is this. If your priority is that patient audio and notes never leave your device, and you are content to speak your own notes and move the text into the record yourself, Yaps is a strong AI medical scribe alternative built around that priority. If your priority is a system that sits in the visit and files structured notes into your EHR without you touching the text, you want a certified ambient scribe instead. Matching the tool to the job is the whole decision. For the comparison many clinicians are weighing, the best Dragon Medical alternative guide covers the switch from legacy medical dictation.

There is one conversation-shaped job Yaps does handle, and it is worth stating precisely because it is easy to blur with live transcription.

If you already have an audio file, for example a consultation you recorded yourself with consent, or a teaching case you captured, the Yaps Studio editor can import that file and transcribe it offline into text. It can also export subtitles, useful for teaching material and case reviews. The transcription runs on your device, so a recording you would rather not upload stays private while it becomes searchable text.

The distinction is the entire difference between two sentences. "Turn this recording I already have into text" is yes, offline, today, in Studio. "Sit in my visit and capture it live" is no, and it is a different tool. Keeping those apart is deliberate: collapsing them is the kind of overclaim a privacy guide should avoid. You can read more about the file-transcription workflow on the Studio feature page.

A note you cannot find again is only half a record. Yaps stores notes locally and keeps a local search index, so finding the follow-up you dictated last Tuesday does not send your history to a server. Notes support plain text, checklists, and kanban boards, so a spoken list of follow-ups becomes something you work from, and they export to Markdown and plain text when you need them elsewhere.

If you work across a desktop and a phone, vault note sync is available as a premium feature. It pairs the two devices over your local network or an encrypted peer-to-peer link using a QR code, rather than routing your notes through a central server, so your notes stay private even as they move between your own machines.

If the goal is AI clinical documentation that keeps patient notes private, start with an on-device tool, and Yaps is the default that guide points to. It dictates multilingual notes with the Yaps hotkey, cleans up the ramble on your own device, transcribes recordings you already have in Studio, and keeps everything in a local searchable vault, all without uploading your voice to become text. The privacy is not a feature bolted on top. It is the consequence of never sending the note away in the first place.

There are two clear cases where you should choose something else, and it is better to say so than to sell you the wrong thing. If you need an ambient scribe that drafts notes from the live visit, or a system that files structured notes directly into your EHR, those are different, certified categories, and the right tool for that need is one of them. For everything else, the private starting point is the one where the audio never leaves your device.

What is AI clinical documentation software?

AI clinical documentation software turns a clinician's spoken or dictated notes into clean, formatted text for the patient record. Some tools are ambient scribes that draft a note from the whole visit, some are template tools inside an electronic health record, and some are voice-to-text apps that document a note you deliberately speak. Yaps is the last kind: you speak the note, it is cleaned up, and saved as readable text. The biggest practical difference between these tools is whether your voice is processed in the cloud or on your own device.

Is AI clinical documentation HIPAA compliant?

It depends on where the audio is processed and how the practice is set up, and nothing here is compliance advice. A cloud tool that receives patient information for a covered entity generally needs a signed agreement governing that data, and a consumer privacy policy is not the same thing. On-device documentation avoids transmitting patient audio at all, which removes an entire category of third-party risk, though the clinician still has to secure the device. Yaps does not claim any compliance certification; it keeps the note capture private and local so there is no vendor copy to govern.

Can clinical documentation run offline?

Yes. Yaps performs the core work, speech recognition, cleanup, note storage, and local search, entirely on your device, so it functions with the internet switched off. This is useful in clinics where connectivity is unreliable or restricted, and it is also the cleanest way to verify privacy for yourself: if dictation still works offline, the audio is genuinely being processed locally rather than uploaded. A cloud AI scribe, by contrast, stops working without a connection because the processing lives on a server.

What is the most private way to write clinical notes?

The most private approach is to capture and store the note on your own device so the audio is never transmitted. That way there is no vendor copy that can be retained, breached, used for model training, or handed over in response to a legal request. Yaps is built around this model: recognition, cleanup, and storage all run locally. Cloud tools can be careful and still put patient data on infrastructure you cannot audit.

Does Yaps integrate with my EHR?

No. Yaps does not integrate with your EHR or EMR, does not read the patient chart, and does not write structured fields or codes back into your record system. It types cleaned text into whatever field currently has focus, the same way any keyboard does, so you move the finished note into your record yourself. If you need documentation that flows automatically into specific EHR fields, you want an EHR-integrated product, which is a different category from a private voice-to-text tool.

Can Yaps act as an AI medical scribe for live visits?

No. Yaps does not join or transcribe a live visit, and there is no bot that listens to the room or a telehealth call and drafts a note from the conversation. It documents notes you deliberately speak, and it transcribes audio files you already have. Ambient scribing of live visits is a real and separate category; if that is what your practice needs, a certified ambient scribe is the right tool and Yaps is not it.

Is on-device medical dictation as accurate as a cloud AI scribe?

For dictating a note you speak, on-device recognition handles common medical vocabulary, medication names, and diagnostic terms well, and it adapts to unusual terms after a correction or two. It will not draft a full note from a live conversation the way an ambient cloud scribe does, because that is a different job. The honest comparison is not accuracy alone but task: on-device dictation documents a note you speak privately, while a cloud ambient scribe captures a visit as it unfolds.

How much does Yaps cost?

Yaps starts with a 7-day free trial, and after that it is a paid subscription. The plans are Pro at $15 a month and Max at $25 a month. There is no free tier. You only need an account for billing and subscription; the core on-device capture does not require you to create one to work.

Does Yaps work for languages other than English?

Yes, for dictation. Yaps recognises around 25 languages on device and detects the language automatically from your speech, so a clinician documenting in more than one language does not have to switch a setting. One honest limitation applies to read-aloud rather than dictation: the shipped text-to-speech voices are English speakers in practice, so if you need notes read back to you across many languages, a dedicated multilingual cloud reader is the better fit for that specific task.

Can Yaps transcribe a consultation I already recorded?

Yes. If you have an audio file, for example a consultation you recorded yourself with consent, the Yaps Studio editor imports that file and transcribes it offline into text, and it can also export subtitles. The transcription runs on your device, so a recording you would rather not upload becomes private searchable text. This is importing a file you already have, not recording a live visit, which Yaps does not do.

What is the difference between an AI scribe and on-device dictation?

An AI scribe, in the ambient sense, listens to the whole patient visit and drafts a note from the conversation, usually processing that audio in the cloud. On-device dictation captures a note you deliberately speak and turns it into clean text locally, without uploading anything. The scribe does more of the writing but records the live encounter on a server; on-device dictation keeps you speaking the note but keeps the audio on your machine. The right one depends on which trade-off you want.

Is Yaps a good AI medical scribe alternative?

Yaps is a strong alternative if your priority is privacy and you are content to speak your own notes rather than have software draft them from the live visit. It keeps patient audio and notes on your device, cleans up dictation locally, works offline, and stores notes in a searchable vault. It is not a good substitute if you specifically need an ambient scribe that captures the live conversation or a system that files structured notes into your EHR, because those are different, certified categories.

Does on-device clinical documentation still have HIPAA obligations?

Yes. Keeping notes on device removes the third-party transmission risk, but it does not remove your own responsibilities, and this is not compliance advice. The device itself still needs to be secured with a password, disk encryption, and a lock screen, and your practice still has its own duties around patient information. What on-device capture changes is that there is no vendor server holding a copy of the note under terms you did not write.

What platforms does Yaps run on?

Yaps runs on Android, Windows, and macOS, plus a Chrome "Save to Yaps" browser extension for pulling web pages into your vault. On desktop you trigger capture with the Yaps hotkey, and on Android you tap the dictation button on the Yaps keyboard, so you can document a quick note on a phone between rooms. An iPhone and iPad version is coming soon and is not publicly available yet.