SpokenroomPre-release

Session notes that never leave your computer.

Spokenroom records your session, transcribes it as you go, separates you from your client, and writes the note — all as processes running on your own Windows PC. Stop recording and go straight into your next appointment; the note is waiting when you are done. There is no server to upload to. No account. No monthly fee.

$59 once.Not $40–120 a month.

Built by a small independent team. No servers, and no investors asking us to monetize your recordings. Early-bird reservations are fully refundable until the release.

The other local tools

Two other tools also keep your sessions on your own computer.

Here is how we differ — and when you should pick one of them instead. Cells where they are better are in bold. "Not stated" means the vendor's site did not say, not that the feature is absent.

 SpokenroomScribularConfidant
Price$59 one-time (1 year of updates; $29/yr to keep updating, optional)$99 one-time (labelled early-access pricing, stated to increase after launch)$359 one-time (1 year of updates; $179/yr optional) or $29.99/month ($24.99/month billed annually)
Free trial7 days, no card, no account14 days, no card14 days, no card
Devices per license22not stated
PlatformsWindows 10/11 (x64, AVX2 CPU) · macOS in v1.5Mac + WindowsmacOS 13+ Apple Silicon only · Windows 11
Minimum memory8 GB (transcript + plain summary) · 16 GB for clinical templatesnot stated8 GB (Lite model) · 16 GB recommended (Pro model)
Where audio is processedYour computerYour computerYour computer
Telehealth recordingYes — microphone + system audio captured as two separate tracksYes — captures Zoom / Meet / Teams / SimplePractice / VSee locallynot stated
Who-said-what in telehealthDecided by track at record time — no acoustic guess, no modelSpeaker tagging (therapist / client / custom names)not stated
Transcription during the sessionYes — transcript is ready when you stopNotes within the session or immediately afterNot stated; generation after recording: up to one minute (16 GB) / closer to four minutes (8 GB)
Note formatsSOAP · DAP · BIRP · GIRP · custom Markdown templateSOAP · DAP · BIRP · GIRPSOAP · DAP · BIRP · GIRP · PIRP · SIRP · custom prompts
Guard against invented contentTwo-stage: evidence first, then prose. Every claim cites a transcript line; quotes must be exact substrings; names, medications, dates and numbers must appear in the cited line; an empty field is written N/A by code, never by the modelnot statedSentence-level check: any sentence whose words cannot be found in the recap is refused
Dictation mode (no recording)v1.5Yesnot stated
RuntimeNative app + signed sidecar processes; no Python, no Ollama; installer ≤150 MBwhisper.cpp + bundled Ollamanot stated
Published benchmarksYes — raw CPU-only numbers, old and new hardwareNoNo
Other verticalsGeneral edition $39; legal edition v1.5Therapists onlyTherapists only
Refunds30 daysnot stated30 days
Who makes itSmall independent teamIndependent developer, Seattle, WAFounded by a licensed therapist (LCSW-R, 16 years in practice); self-funded

Scribular and Confidant figures from their public websites, checked 2026-10-18. We recheck quarterly; if you find a row out of date, email us and we will fix it the same week.

Full comparison, including when to choose them instead →

What it produces

A transcript, and the note that comes out of it.

Below is a pair from our test suite: a fictional session written for testing, and the reference note we hold the software's output to. This is the format, the field discipline, and the standard — measured model output replaces it at release, including the cases where it falls short.

Transcript · telehealth sessionmicsystem
00:00:04ClinicianGood to see you. How has the week been since we last talked?
00:00:09ClientHonestly? Rough. I think I slept maybe four hours a night most nights. I keep lying there running through the presentation over and over.
00:00:24ClientIt's Thursday. And every time I think about standing up in front of the whole leadership team my chest just gets tight. Like actually physically tight.
00:00:36ClinicianTell me more about what that's like in your body.
00:00:40ClientIt's like someone's sitting on my ribs. And my hands get cold and clammy. I had to leave a meeting on Tuesday because I thought I was going to be sick.
00:01:42ClientI did it twice. The first time it felt stupid, like I was just arguing with myself on paper. But the second time I wrote down "I will freeze and forget everything" and then the evidence column and I realized I've given four of these already and I've never once frozen.
00:02:08ClientKind of annoying, actually. Because the feeling didn't change.
Note · SOAP · referenceLocal · 1.7B / 4B

Subjective

  • Client reported sleeping approximately four hours per night for most of the past week, attributing the insomnia to repeatedly rehearsing an upcoming quarterly presentation scheduled for Thursday.
  • Client described chest tightness when anticipating presenting to the leadership team: "like someone's sitting on my ribs," with cold, clammy hands.
  • Client reported leaving a meeting on Tuesday due to nausea, remaining in the bathroom approximately twenty minutes until symptoms passed.
  • Client reported completing the thought record twice, identifying four prior presentations delivered without freezing, while noting the anxious feeling did not change in response to that evidence.

Objective

  • Client was oriented and engaged throughout and participated in an in-session breathing exercise.
  • Client voiced mild skepticism about the thought record while continuing to engage with the task.

Risk

N/A — not discussed in this session

Assessment

Session focused on anticipatory anxiety regarding a work presentation, with somatic symptoms consistent with a panic episode reported on Tuesday. The previously identified belief "I don't belong" recurred and remains active. Client is completing between-session work and able to generate disconfirming evidence cognitively while reporting that affect has not yet followed.

Plan

  • Continue thought record when the "I don't belong" thought arises.
  • Practice 4-4-6 breathing, three rounds twice daily, before Thursday.
  • Behavioral experiment: stop rehearsal at 9:00 PM Wednesday and report the outcome.
  • Session moved to Friday 4:00 PM to review the presentation.
It writes N/A

When a field has no basis in the transcript, the note says so instead of filling the space. A complete-looking invented note is a clinical hazard; a short accurate one is correct.

Quotes are real quotes

Every quotation is checked against the transcript — and against the right speaker — before it reaches you. Not a request we make of the model, a check the software performs and a claim it will drop.

No diagnosis

The note will not state, suggest, or imply a diagnosis unless a diagnosis was actually spoken in the session, and then it is attributed.

We assume the model will make things up, and we built the software to catch it.

Early in development we ran a small language model against a test session and asked it for a SOAP note. For the Objective field — where that transcript genuinely contained nothing to report — it invented three observations and put six phrases in quotation marks that appear nowhere in the recording. It would rather write a complete-looking paragraph than leave a field empty. Every model we have tested behaves this way to some degree.

So the note is not written in one pass. The model first has to extract candidate facts and cite the transcript line each one came from. The software then checks every citation, every quotation, and every name, dose, date and number against that line, and against the right speaker. Anything that fails the check is dropped rather than softened. If nothing survives for a field, the field reads "not discussed in this session" — and that decision is made by the checker, not by the model.

Two things we learned building this, which is also why clinical templates need 16 GB. The smaller model that fits in 8 GB attributed the clinician's own words to the client, in quotation marks — the larger one got all four attributions right and paraphrased instead of borrowing. But both of them invented an Objective section, so a bigger model is not the fix; the checker is. We would rather tell you that than imply the output is trustworthy because the model is clever.

Why this exists

You already know what's wrong with the current options.

You pay $40–120 a month for notes, and the bill arrives whether you wrote ten or a hundred. At least one popular tool starts billing when a session is merely scheduled, before any note exists.

Free tiers exist because the session trains the model. If you explained that trade to your client in plain language, would they recognize it as the arrangement they consented to?

The vendor says "HIPAA compliant," but the signed agreement is on the enterprise plan, and an agreement is a promise rather than an architecture. Meanwhile the recording sits in someone else's cloud: discoverable, breachable, and outside your control.

There is a simpler answer available in 2026 that was not available in 2022. Don't send the audio anywhere. The models are now small enough and fast enough to run on the computer already on your desk.

How it works

Four steps, none of which involve the internet.

  1. Press record.

    Microphone only for an in-person session. Microphone plus system audio for telehealth.

  2. Have the session.

    Spokenroom transcribes as it records, quietly and at low priority, so it never competes with your video call. By the time you stop, the transcript is written.

  3. Pick a template.

    SOAP, DAP, BIRP, GIRP, or your own written in plain Markdown. The note follows your fields exactly.

  4. Go to your next session.

    The note is queued, not something you wait for. Spokenroom writes it between appointments and at the end of the day, so you review a stack of finished notes instead of sitting through a progress bar six times. One click copies a note to the clipboard; DOCX and PDF export too. The audio is deleted the moment processing finishes — that is the default, not an option you have to find.

On telehealth, speaker separation costs nothing — because it isn't guessed.

Your microphone records you. Your system audio records your client. Spokenroom keeps those as two separate tracks rather than mixing them, so who said what is a fact established at recording time, not a prediction made afterwards. That removes the single most common complaint about tools in this category: a sentence split between the wrong speakers. In-person sessions have only one microphone, so there we do identify speakers acoustically — and we show you the transcript and note first, rather than making you wait for it.

Built for back-to-back sessions, because that is how the day actually runs.

Writing a progress note by hand takes about half an hour. A tool that hands it to you in sixty seconds but requires you to sit and watch has solved the wrong problem, because at the end of a fifty-minute hour the next client is already waiting. So notes go into a queue: recording stops, you move on, and the queue works through itself between appointments. When you sit down to do your documentation, it is already written and waiting for your review and your signature.

Comparison

The differences that actually matter.

 SpokenroomTypical cloud note tool
Where audio is processedYour computerVendor's servers
Price$59 once$20–120 / month
Billed per noteNeverPer note or per session on most plans
Works offlineFullyNo
Account requiredNoYes
Signed data agreement neededNo — nobody receives the dataYes, often enterprise-only
Your sessions train a modelNoSometimes, on free tiers
If the company shuts downYour copy keeps workingThe service stops

Competitor pricing ranges are from public pricing pages, checked 2026-10-08. The Mentalyc comparison names figures individually.

Privacy

"Private" should mean something you can check.

Not a policy page. A list of the connections the software makes.

Three outbound calls, each with a switch

License check, which also works offline. Update check. Model download, only when you ask for one. That is the complete list.

No analytics, no telemetry

There is no analytics SDK in the application. Crash reports are opt-in, anonymous, and contain a stack trace and nothing else.

Your folder, your disk

Audio, transcripts and notes live in a directory you choose. Point it at an encrypted volume if your policy calls for that.

Full Offline Mode

One switch blocks every outbound connection, and an indicator shows when it is on. The product is designed to be fully usable in that state.

Audio deleted by default

When processing finishes, the recording is overwritten and removed unless you have chosen to keep it.

A consent template included

An editable, printable client-facing explanation of what the software does, so you can document the conversation rather than improvise it.

Requirements

It runs on the laptop you already have.

Within reason, and we would rather tell you the limits now than refund you later.

 
Runs
Clinical notes
Operating system
Windows 10 or 11, 64-bit
Windows 10 or 11, 64-bit
Processor
4 cores with AVX2 (roughly 2013 onward)
8 cores recommended
Memory
8 GB
16 GB
Disk
5 GB
8 GB
Graphics card
Not required
Not required; a Vulkan GPU makes notes faster and enables non-English transcription
SOAP, DAP, BIRP and GIRP need 16 GB of memory. Read this before you buy.

On 8 GB we have to run a smaller language model, and in our testing that model attributed the clinician's own words to the client — in quotation marks. The larger model that needs 16 GB got every attribution right on the same transcript. Putting the clinician's words in the client's mouth is a factual error in a medical record, not a wording preference, so we do not ship clinical templates on the 8 GB tier at all. An 8 GB machine still records, transcribes, separates speakers and produces plain summaries. If your laptop has 8 GB and you need SOAP notes, this is not the right purchase yet, and we would rather lose the sale than take it.

Full requirements and how we measure →

Pricing

One price, paid once.

Spokenroom for Therapists
$49early-bird, then $59 at release
  • Unlimited sessions, unlimited minutes, forever
  • One year of updates included; after that your version keeps working
  • Optional update renewal at $29/year, never required
  • Two machines per licence — office desktop and laptop
  • Seven-day trial at release: every feature, no card, no account
  • Refundable in full until the release, and for 30 days after

Clinical note templates require 16 GB of memory. See requirements — we would rather you check first than refund later.

Reserve at $49

Checkout opens when the merchant-of-record account is approved; until then, reserving puts you on the list for the early-bird price. Payment and invoicing are handled by our merchant of record, who also handles VAT and sales tax. We never see your card.

Also available
$39general edition
  • Same engine, without clinical templates
  • Interviews, meetings, research, subtitles
  • Upgrade to the therapist edition for $20
General edition →
Founding Clinician programme

The first 200 clinicians get it free, for life.

I need to know whether these notes are genuinely good enough for real clinical work, and the only people who can tell me that are clinicians.

Use it on five real sessions, answer six honest questions, and the licence is yours permanently — including every future major version. If your answer is that it is not good enough, that is the single most useful thing you could do for this project.

There is no catch and no obligation to say anything kind.

Or just keep me posted

Double opt-in. No tracking pixels in the emails. Unsubscribe in one click.

Questions

The ones we actually get asked.

Does it need an internet connection?
No. After the one-time model download, Spokenroom works with Wi-Fi switched off. Licence activation also works offline.
Is my client's audio uploaded anywhere?
No. Transcription and note generation run as local processes on your computer. There is no upload step in the product.
Do I need a signed data agreement with you?
No. That agreement exists because a vendor handles protected health information. We never receive it, so there is no relationship to paper. If your employer requires the paperwork regardless of architecture, a cloud tool may suit your situation better, and we would rather say so than sell you the wrong thing.
Is this HIPAA compliant?
Compliance is a property of your practice, not of a product, and any vendor who tells you otherwise is selling. What we can state precisely: Spokenroom does not transmit, store, or process your session data on any system we control. Keeping the data on a device you already secure is the configuration most clinicians are already required to maintain.
How accurate are the notes?
The note is generated from the transcript by a local language model, constrained to your template, and every claim is checked against the transcript line it came from before it reaches you. It fills fields; it does not compose from imagination. You review and edit before anything enters a chart, exactly as with any scribe. Our beta quality scores will be published, including the cases where the model performed badly.
Will it suggest a diagnosis?
No. It will not state or imply a diagnosis unless one was spoken in the session, and then it attributes it rather than asserting it.
What happens when a client mentions self-harm?
The risk field quotes the speaker verbatim and flags the note for your review. The software does not summarise, soften, or rate risk, and it never concludes that risk is absent — if nothing was said, the field reads "not discussed in this session." That judgement is yours and the design keeps it there.
Does it integrate with my EHR?
Not directly. One click copies the finished note so you can paste it into any system. A direct integration would mean routing your data through us, which is the thing we are avoiding.
What if I only have 8 GB of memory?
Recording, transcription, speaker separation and plain summaries all work. Clinical templates — SOAP, DAP, BIRP, GIRP — do not: the smaller model that fits in 8 GB made attribution errors in our testing, assigning the clinician's words to the client, and that belongs in a record even less than a short note does. Spokenroom detects this during setup and tells you, and the requirements page says it before you pay.
What happens if I stop paying?
Nothing, because there is nothing to stop paying. The $59 includes a year of updates; afterwards you keep the version you have, permanently, and can renew updates for $29 if you want them.
You're a small team. What if you disappear?
The software runs locally and does not check in with us in order to function. If we vanish, your copy keeps working — which is not true of any cloud tool you are currently paying for.