There is a version of the Wispr Flow privacy argument that runs on impressions: the app feels private, the settings page has switches, and the company says it does not sell data. Then there is the version that runs on the vendor's own documentation. The second version is one sentence long.
\n\n\n“Transcription always occurs on the cloud. This is the best way for us to provide accurate, low latency transcription.”
(The product was called Whisper Flow until the rename; whisperflow.com redirects to wisprflow.ai, and the documentation below is the same vendor's.)
\nThat is Wispr Flow's Data Controls page, not a critic's summary of it. The company is being unusually direct, and it deserves credit for saying it plainly instead of hiding behind the word \"secure\". It also settles a question that a lot of people get wrong about this category: the model that turns your voice into text does not run on the device in your hand.
\n\nPress the dictation key and the following happens, every time, on every platform:
\nNone of that is a scandal. It is how most dictation products work, and it is why they are fast on hardware that could never run the model itself. But it has four consequences that no setting can undo, because they are properties of the architecture rather than of the policy:
\n| Claim | What it actually promises | Wispr Flow | Jot |
|---|---|---|---|
| Transport | Audio is encrypted on the way to a server | Yes | Not applicable — no upload |
| Rest | The server deletes it on a schedule you configure | Yes, configurable | Not applicable — no server copy |
| Absence | The audio never leaves the machine | No | Yes |
Only the third survives a determined adversary, a subpoena, or a change of ownership. The first two are policy commitments, and policy is exactly as strong as the vendor's controls on the day you need them.
\n\nJot runs its speech model on the phone. The audio is decoded locally, the text appears locally, and the recording is never uploaded — there is no server-side copy to retain, delete, or leak. The App Store listing states the architecture in the same blunt register Wispr uses: “No text is saved, transcription is done on device, corrective AI calls are de-identified, use American ZDR providers, and optional.”
\nThat last word matters. Jot does make network calls, but only for an optional corrective pass, and only with de-identified text — never audio — to providers under zero data retention agreements. Turn the feature off and the app is a closed system. That is a different bargain from \"we upload your voice but promise to delete it\", and it is the bargain most people think they are getting when they read the word \"private\".
\n\nAccuracy on hard audio. Overlapping speakers, heavy accents, a bad microphone in a noisy room — a frontier model on a server GPU beats anything that fits in a phone's memory budget, and it does so without heating the device or draining the battery. Wispr's speed is real, and it is bought with the upload.
\nIf your recordings are ordinary and your threat model does not include the vendor, that trade is fine. If the recording is a therapy session, a medical intake, a deposition, or a board meeting, the trade is the whole question.
\n\nWispr Flow is cloud transcription with a good privacy policy. The policy is a promise about a server that holds your audio; the architecture is what puts it there. You can improve the promise in settings. You cannot change the architecture in settings.
\nIf you need dictation to work without a connection — on a plane, in a basement, in a building that blocks outbound audio — cloud transcription is disqualified before privacy enters the conversation.
\nIf you want the audio to stay on the phone, that is a different product, and Jot on the App Store is the one this site is about.
\nWhether Wispr Flow keeps your words is not one question. It is at least four, and the answers differ by feature. The documentation separates them cleanly enough that the whole post can be built out of quotations.
\n\n\n“Dictation Cloud Storage controls whether Wispr stores your transcription data (transcripts, audio, and dictation history) on Wispr's servers. Dictation Cloud Storage enables current and future features that require persistent, cross-device access to your data.”
Read that second sentence carefully, because it is doing real work. Cloud storage is not a side effect of the service — it is the foundation of cross-device features. That means the incentive runs one way: as long as the product roadmap adds features that need a server-side history, the default will drift toward keeping one.
\nWispr also offers the opposite control, and it is genuinely good:
\n\n“You choose whether dictation history is kept on your computer: keep it, auto-delete every 24 hours, or never store it. That one is desktop today, with mobile coming soon.”
Three options, including never. That is a better retention story than most of this category. The catch is in the last eight words: on a phone, that control is not there yet.
\n\n\n“Notetaker transcripts are stored on Wispr's private cloud for cross-device syncing and resumption of meetings.”
A meeting recorder that syncs across devices has to store the meeting somewhere. Wispr says so, and says where. Meeting audio is handled differently:
\n\n“Wispr may temporarily retain encrypted meeting audio on a user's device and/or in cloud storage for a limited period to support requested Notetaker functionality, including allowing users to resume meetings, improve transcription quality, verify transcription accuracy, troubleshoot the Services, and provide related functionality.”
\"Temporarily\" and \"a limited period\" are honest but unfixed. The retention window is whatever the service is configured for, not a number you can hold a vendor to from outside. And note the phrase \"on a user's device and/or in cloud storage\" — the audio can be in both places at once.
\n\n| Surface | Where it lives | What you control |
|---|---|---|
| Dictation transcript and history | Wispr servers, if Dictation Cloud Storage is on | Cloud storage on or off |
| Dictation audio | Wispr servers, same switch | Cloud storage on or off |
| Local dictation history | Your computer | Keep, 24 hours, or never — desktop only |
| Notetaker transcript | Wispr private cloud, for sync and resumption | Notetaker retention setting |
| Notetaker meeting audio | Device and/or cloud, encrypted, \"a limited period\" | Not published as a fixed number |
| Context from the active window | Processed by the service when Context Awareness is on | Context Awareness on or off |
Wispr Flow ships on iOS and Android, and mobile is where a dictation tool is most useful — walking, commuting, between meetings, one thumb on a keyboard. It is also the platform where a phone is most likely to be lost, shared, borrowed, or imaged by someone else, and the platform where \"keep it on your computer\" is not available yet.
\nSo the honest summary is not that Wispr Flow retains everything. It is that the strongest retention control it publishes is a desktop control, and the mobile product — the one most people actually use — is still waiting for it.
\n\nA retention setting is a promise that a copy will be deleted. Local-only storage is the absence of the copy. If the transcript never leaves the phone, then a breach at the vendor has nothing of yours in it, a subpoena to the vendor returns nothing, and an acquisition does not transfer your history as an asset.
\nJot keeps drafts and transcripts on the device, transcribes on the device, and uploads no audio at all. The App Store listing puts it as plainly as Wispr puts its own architecture: “No text is saved, transcription is done on device.” The one network path is the optional corrective pass, which sends de-identified text — not audio — to providers under zero data retention agreements.
\n\nWispr Flow gives you real control over dictation retention. Off, or 24 hours, is a legitimate answer, and on desktop it is one toggle away. If you are on a Mac and you like the product, use it.
\nIt does not give you local storage. Cloud storage is the substrate for the features the company sells, Notetaker is stored in a private cloud by design, and the local-history control has not reached mobile.
\nThe distinction that matters: a setting can be changed by a vendor, a roadmap, or an acquisition. An architecture cannot. If you want the copy never to exist, Jot on the App Store is the local option.
\n\"Zero data retention\" is the most over-read phrase in privacy software. It sounds like a property of the product. In almost every case it is a property of a contract between two companies, one of which you have never heard of and both of which already have your data.
\n\n\n“Wispr always maintains zero data retention agreements with all third-party AI providers. All third party AI models used in Flow Dictation are subject to these agreements. Certain features, such as briefs for Notetaker, rely on features from third party AI providers that are not supported under zero data retention.”
Three sentences, three distinct claims, and the third one is the interesting one. The company is telling you, in its own documentation, that the zero-retention guarantee has an exception. Notetaker briefs are processed by providers that cannot offer ZDR. Wispr does not say what happens to that data instead; it says only that those providers \"cannot use your data for model training\".
\nThat is a meaningful protection. It is not zero retention.
\n\n\n“Customer content is also processed by third-party AI and language model providers to deliver the service. They process it solely to provide the service, and it's generally deleted within 30 days.”
Two words in that sentence are doing a lot of work: generally, and 30 days. A zero-day promise and a 30-day norm are not the same commitment, and they are published by the same company about the same pipeline. The most generous reading is that the two sentences describe different features. The least generous reading is that they describe the same one. Either way, a buyer cannot resolve it from outside, and that is precisely the problem with promises made about servers you do not control.
\n\nVendors use the phrase for at least three materially different positions, and the badges are not interchangeable.
\nThe transcription vendor keeps your data for the length of the job and deletes it immediately after, with a contractual zero-day policy flowing down the chain. This is a real and meaningful guarantee — but it only describes the seconds-to-minutes window inside a vendor you have already handed the audio to.
\nYour content is never used to improve a model. This is the single most load-bearing privacy clause in the category, and the strongest version names third parties explicitly, so the promise survives the vendor swapping model providers.
\nNothing is retained because nothing is transmitted. There is no copy on anyone's infrastructure, which means there is no retention window to configure, no DPA to negotiate, and no breach of yours to worry about on the vendor's side.
\nMost products in this comparison are at position 1 or 2. Position 3 is a different architectural claim, not a stronger version of the same one.
\n\n| Layer | Covered by a ZDR claim? | What it means for you |
|---|---|---|
| Third-party model provider (Flow Dictation) | Yes, per Wispr | Your audio is not kept by that provider |
| Third-party model provider (Notetaker briefs) | No, explicitly excluded | Not covered by the ZDR commitment |
| Wispr Flow itself | No | Governed by your retention setting, not by ZDR |
| Your device | Not applicable | Audio is uploaded before transcription happens |
| Jot | Not applicable | Audio is never transmitted, so there is no retention window to negotiate |
Every row except the last one is a promise about someone else's database. The last row is a statement about where the audio is, which is the only kind of statement that does not need to be believed.
\n\nBecause it is a good contract term and a better marketing term. ZDR agreements are genuinely valuable: they stop a subprocessor from keeping training corpora built out of your meetings, and they are the difference between a vendor that thought about the supply chain and one that did not. The mistake is reading a supply-chain guarantee as a product guarantee.
\nA product-level version of the claim exists, and it looks different. Jot's App Store listing: “No text is saved, transcription is done on device, corrective AI calls are de-identified, use American ZDR providers, and optional.” The ZDR part applies to the optional network calls. The \"no text is saved\" part is true because of where the work happens, not because of what a supplier promised.
\n\nWispr Flow's ZDR claim is real and narrower than it sounds. It covers third-party model providers for dictation, excludes at least one feature family, and says nothing about how long Wispr itself keeps your content.
\nEnterprise customers get the strongest version — data sharing off, no option to enable it — but that is a contract, not an architecture, and it costs a procurement cycle.
\nIf you want retention that does not depend on a supplier's contract, the audio has to stay where you can see it. Jot on the App Store.
\nThe most common mistake people make with transcription privacy settings is assuming there is one switch. There are usually three, they control different things, and the vendor's documentation is the only reliable place to learn which is which.
\n\n\n“When you rate a dictation or submit feedback through a form or by clicking thumbs up or thumbs down, we collect your feedback and also data from the related session. Session data may include audio recordings, transcripts, your edits, the app where you dictated, and technical information such as device, app version, and performance data. We use this information to diagnose issues, improve Wispr Flow, and develop and train our AI models.”
That is Wispr Flow's Data Controls page. It is a clean, honest paragraph — and it is the single most important one on the page, because it describes a path by which audio recordings reach model training even though the headline training setting is about text.
\nClicking a thumbs-down is an ordinary thing to do. It is also, per that paragraph, an act that attaches a recording of the session to a feedback record the company may use to develop and train models.
\n\n\n“Whether Customer Content is synchronized across devices or retained to provide Service functionality is controlled by the applicable Service features, not by your model training preference.”
Wispr says this outright, and it is the sentence to keep. Turning off model improvement does not turn off cloud storage. Turning off cloud storage does not turn off the feedback path. The controls are orthogonal, and only one of them is about training.
\n\n\n“If you opt to share your content with us for model training, we may also collect pseudonymized text, meeting transcripts, meeting outputs, and corrections you provide to improve the performance of Wispr Flow for all users.”
Two things are worth noticing. First, the word is pseudonymized, not anonymized — the two are not synonyms, and only one of them means the data cannot be traced back. Second, the setting is described as something you opt into. Wispr's documentation does not state, on the pages cited here, what the default is for a new account, and a default is exactly the kind of thing a buyer should be able to read in one line rather than infer.
\nEnterprise is the exception, and it is unambiguous:
\n\n“Enterprise accounts have data sharing turned off, and there is no option to enable it.”
So the strongest training guarantee in the product is a contract tier. Everyone else is trusting a switch whose default the documentation does not pin down.
\n\n| Switch | What it controls | What it does not control |
|---|---|---|
| Model improvement | Whether your content may be used to train or improve models | Storage, sync, or the feedback path |
| Dictation Cloud Storage | Whether transcripts, audio and history sit on Wispr servers | Training, or the subprocessor chain |
| Feedback (thumbs, forms) | Nothing — it is an action, not a setting | Attaches session audio and edits to a record |
| Enterprise contract | Data sharing off, no user-level override | Requires a sales process and a budget |
Jot's App Store listing describes a narrower design: transcription on the device, no text saved to a server, and an optional corrective pass that sends de-identified text to US providers under zero retention agreements. There is no corpus of your dictation to train on, because the dictation is not sent anywhere in the first place.
\nThat is not a stronger promise. It is a smaller surface. A product that never receives your audio cannot train on it, cannot lose it, and cannot change its mind about it.
\n\nYes, Wispr Flow can train on your audio — through the feedback path, per its own Data Controls page, regardless of how you feel about the model-improvement toggle.
\nThe settings are honest and separate. Wispr documents all of this, which is more than most competitors do. The problem is not disclosure; it is that a cloud product needs three switches to approximate what local processing gets by default.
\nIf you do not want a training corpus to be possible, the answer is not a better switch. It is Jot on the App Store.
\nMost privacy discussions about dictation stop at the microphone. That is the wrong place to stop. A dictation tool is a keyboard replacement, which means it sits inside whatever app you are using, at the moment you are using it — and the text around your cursor is often more sensitive than the words you say out loud.
\n\n\n“If you enable the optional Context Awareness feature, we may collect limited, relevant content from the specific app in use (such as the text on the screen) to enhance the accuracy of Wispr Flow's Outputs.”
Opt-in, and described accurately. If you turn Context Awareness on, expect the service to read text from the active window. That is the deal, it is disclosed, and it can be turned off. Nothing unfair is happening.
\n\n\n“Wispr always uses information about the app you are dictating in (e.g. app name) to format messages (e.g. formal tone in email, casual tone in messages). It also takes into account the surrounding textbox content to help determine correct capitalization and punctuation, and to edit text with Command Mode.”
The word is always. App identity and surrounding textbox content are part of the product, not a feature you enable. It is easy to see why: tone selection and punctuation repair genuinely need context, and the results are better for it.
\nIt also means that during a dictation, the service sees the field you are typing into. A password reset form, a medical portal, a support ticket containing someone else's data, a draft you have not sent — all of it is \"surrounding textbox content\" by that definition.
\n\nCloud dictation already sends your voice. Context awareness sends the neighbourhood. Together they reconstruct something close to \"what you were doing at 2:14pm\", which is a different and more useful category of data than either alone.
\nWispr's answer to this is disclosure plus switches plus an Enterprise tier with data sharing off. That is a reasonable answer from a company that has to run a cloud service. It is not the same as a product where the text never leaves the phone.
\n\nJot transcribes on the device and inserts into the field you are in, which means the surrounding text is on the same machine it was already on. There is no upload of the window, because there is no upload of the dictation. The optional corrective pass sends a de-identified text snippet — never audio, never app identity — to a zero-retention provider, and it can be turned off entirely.
\nThat is the structural difference in one sentence: Wispr needs context on a server to format your text; Jot has the context already, because the text is being produced there.
\n\nContext Awareness is opt-in and disclosed. If you turn it on, you know what you are getting, and the accuracy gains are real.
\nApp awareness and textbox context are not opt-in. They are part of the formatting engine, and they mean the service sees more than your voice.
\nIf the surrounding text is the sensitive part — and for anyone dictating into a clinical, legal, or financial system it usually is — Jot on the App Store keeps that context on the device.
\nThere is one test that settles this, and it takes less time than reading a privacy policy: put the phone in airplane mode and dictate a sentence. If text appears, the model is on the phone. If nothing appears, the model is on a server and the app has been telling you about its policy, not its architecture.
\nEverything else in this post is detail around that test.
\n\nA laptop can hold a large model, run it on a fan-cooled chip, and stay plugged in. A phone has a battery, no fan, and a thermal budget measured in seconds of sustained load. That is why almost every transcription app you have heard of runs the model on a server and uses the phone as a microphone with a screen.
\nThe trade is legible: cloud models are more accurate on difficult audio and much faster to improve, because the vendor ships a new model by changing a server. On-device models are fixed at install time, limited by the device, and cannot be improved for you without an app update.
\nWhat you get for that trade is the thing the cloud cannot offer: the audio never leaves the device.
\n\nJot is a keyboard-first transcription app for iOS 26.0 or later, on iPhone, iPad and iPod touch. It transcribes speech on the device, inserts the result where you are typing, and keeps drafts and transcripts locally.
\nThe App Store listing states the architecture directly: “No text is saved, transcription is done on device, corrective AI calls are de-identified, use American ZDR providers, and optional.”
\nThe optional part is the corrective pass. When you turn it on, Jot can send a de-identified text snippet to a US language-model provider under a zero data retention agreement — no audio, no account, no identifier. Turn it off and the app makes no transcription calls at all. It is a single purchase, $49.99 once, rather than a monthly seat.
\n\n| App | Where the model runs | Works offline | Price model |
|---|---|---|---|
| Jot | On the iPhone | Yes | $49.99 once |
| Wispr Flow | Cloud — “transcription always occurs on the cloud” | No | $144 a year (Pro, annual) |
| Otter.ai | Cloud | No | $99.99 a year (Pro, annual) |
| Fireflies.ai | Cloud | No | $120 a year (Pro, annual) |
| tl;dv | Cloud | No | $216 a year (Pro, annual) |
The prices are the vendors' published annual plans, and they are per user. Every one of the four cloud products above costs more in its first six months than Jot costs ever.
\n\n\n\nOn heavily accented speech, in a noisy room, with overlapping speakers, a frontier cloud model still wins. Jot's on-device model is good at the job most people actually have — one person, one microphone, ordinary acoustics — and it is not going to beat a server GPU at the hardest 5% of recordings.
\nWhat it does beat is the assumption that every recording has to be uploaded. Most do not, and the ones that do are usually the ones you would least want on someone else's disk.
\n\nRun the airplane-mode test on whatever you use. If it fails, the app is a microphone with a server behind it, and its privacy claims are contract terms rather than architecture.
\nIf it passes, ask what the network is still used for. Jot uses it only for an optional, de-identified corrective pass that you can switch off.
\nIf you want the iPhone to do the transcription, Jot on the App Store — $49.99 once, no subscription.
\nComparing two dictation apps on a desktop is mostly a comparison of features. Comparing them on a phone is a comparison of architectures, because the phone is the device with the least room for a model and the most to lose from a lost or borrowed device.
\nWispr Flow and Jot both ship on iOS. They do not do the same thing with your voice.
\n\nWispr Flow's Data Controls page states that “transcription always occurs on the cloud”. Jot's App Store listing states that “transcription is done on device”. Every other difference in this post follows from those two sentences.
\n\n| Jot | Wispr Flow | |
|---|---|---|
| Speech model runs on | The iPhone | Wispr's servers |
| Works offline | Yes | No |
| Audio uploaded | Never | For every dictation |
| Network used for | Optional, de-identified corrective pass | Transcription itself |
| Transcript stored | On the device | On Wispr servers if cloud storage is on; Notetaker in its private cloud |
| Local history control on mobile | Local by default | Desktop today, “with mobile coming soon” |
| Price model | $49.99 once | $144 a year (Pro, annual) |
| Free tier | — | 1,000 words a week on mobile |
| Notetaker / meeting bot | No | Yes, on Mac and Windows |
It is faster to get started, it has a genuinely useful free tier at 1,000 words a week on mobile, and its formatting is good enough that people stop thinking about it. It also has a team story and a support organisation that a one-person product does not. If your dictation is ordinary and you want the most polished experience on the App Store today, Wispr Flow is a defensible choice.
\nIt also has something Jot does not: a meeting recorder that runs on the desktop and syncs to the phone. If meetings are your use case rather than dictation, that is the deciding feature, and no privacy argument changes it.
\n\nWispr's best retention control is a desktop control. In the company's own words: “you choose whether dictation history is kept on your computer: keep it, auto-delete every 24 hours, or never store it. That one is desktop today, with mobile coming soon.”
\nSo the answer to \"can I keep my dictation history off the cloud on my phone\" is, at the time of writing, no. The audio is uploaded because transcription needs it, and the history control that would keep a local copy from being the only copy has not shipped on mobile.
\nOn Jot there is nothing to configure, because there is no cloud copy to begin with. A lost phone is a lost phone; it is not also a vendor incident.
\n\nWispr Flow Pro is $15 a month billed monthly, or $12 a month billed annually — $144 a year. Jot is $49.99 once, with no renewal.
\nThat comparison assumes one person and one device, which is the situation a phone app is usually bought for.
\n\nCloud-model accuracy on the hardest audio, a meeting recorder, team administration, SOC 2 paperwork for a procurement review, and the fluency that comes from a large engineering team. If any of those are requirements rather than preferences, Wispr Flow is the answer.
\nWhat you get back is a dictation keyboard that works in a basement, does not upload your voice, and does not renew.
\n\nOn architecture, this is not close: Wispr Flow uploads every dictation, and Jot does not. No setting changes either fact.
\nOn polish and scope, Wispr Flow wins. It is a bigger product with a better free tier and a meeting recorder that Jot does not attempt.
\nOn the phone, the question is whether you want a transcript that exists only on the device in your hand. If you do, Jot on the App Store.
\nMost \"best transcription app\" lists are ranked on features, which is the least interesting axis in this category. The features are all good. The differences that matter are architectural, contractual, and commercial, and they are all knowable from the vendors' own pages.
\nFour columns carry the decision:
\n| App | Where transcription runs | Training on your content | Retention | Price model | iPhone |
|---|---|---|---|---|---|
| Jot | On the device | Not applicable — nothing is transmitted | Nothing retained; drafts stay on the device | $49.99 once | App, on-device |
| Wispr Flow | Cloud | A setting you control; feedback pulls in session audio | Local history: keep, 24h or never — desktop only | $144 a year | App, cloud backend |
| Otter.ai | Cloud (AWS) | On by default, automatic | No published figure; Trash clears at 30 days | $99.99 a year | App, cloud backend |
| Fireflies.ai | Cloud, transcription via Assembly.ai | Never | Zero-day at subprocessors; your meetings are stored as the product | $120 a year | App, cloud backend |
| tl;dv | Cloud, EU data residency | Never, contractually | Not published | $216 a year | App, cloud backend |
| MacWhisper | On-device, local models | Not applicable, nothing leaves | Nothing retained, because nothing is transmitted | €64 once | Separate iOS build |
Only two products in that table are bought rather than rented, and only two run the model locally. Jot is the only one that is both.
\n\nA genuinely configurable retention story for dictation — you can turn cloud storage off entirely, or set local deletion at 24 hours. It is also the best-funded product in the category, which is why it feels the most finished.
\nModel improvement is a separate switch from storage, feedback collects audio recordings and edits, and only an Enterprise contract turns data sharing off org-wide. And the architecture is cloud by design: “transcription always occurs on the cloud”.
\nThe deepest meeting-bot integrations and the broadest language support in the category, on the strength of a $100M ARR business.
\nTraining on your audio is automatic and describes itself as such; no user-facing opt-out could be verified, and there is active litigation over exactly this.
\nThe most rigorous security posture in the category, and the only vendor here holding ISO/IEC 42001 for its AI management system. Its trust centre publishes the actual audit artefacts.
\nEvery subprocessor is US-based with no EU data residency option, and custom retention is an Enterprise-only feature — so its zero-retention promise stops at the vendor boundary, not at the meeting.
\nThe cleanest no-training clause in the category — it extends the prohibition to any third party, not just to tl;dv's own models — plus genuine EU data residency for buyers who need it.
\nAll of that protection is defeated in practice by an August 2026 exposure of roughly 181,874 meeting records through a missing database authorisation rule. Residency and contracts do not substitute for correct access control.
\nIt is the only product here that is genuinely local, and the only one that genuinely sells a licence rather than a subscription. On the two axes this comparison measures, it is the harder competitor.
\nNo compliance certifications at all, which rules it out for regulated procurement outright, and macOS is a hard ceiling with no Windows, Linux, or web build.
\nOn-device transcription, local drafts, an optional de-identified corrective pass, and a one-time price. It does not do meetings, teams, admin consoles, or compliance paperwork, and if you need any of those it is the wrong product.
\n\nPrice model is usually treated as a preference. It is a statement about architecture. A vendor that charges monthly has a recurring cost to cover, and in this category that cost is usually inference on a server. A vendor that charges once has to have near-zero marginal cost per use, which is only possible when the model runs on hardware the customer already paid for.
\nSo \"subscription or licence\" is not really a pricing question. It is the same question as \"cloud or local\", asked in a different currency.
\n\nThere is no single best transcription app. There is a best app for a threat model, a platform, and a payment preference, and the table above makes all three visible.
\nThe narrow claim this site makes is that Jot is the only product here that combines on-device processing with a one-time price, and that combination is the reason the rest of these comparisons exist.
\nIf those two properties are what you want, Jot on the App Store.
\nIf you want a transcription app you pay for once, there are two worth considering: Jot at $49.99 and MacWhisper at €64. Everything else that leads this category — Wispr Flow, Otter.ai, Fireflies.ai, tl;dv — is a subscription, at $99.99 to $216 a year.
\nThat is the short answer. The rest of this post is the arithmetic, the reason the category is priced this way, and how to tell a genuine licence from a \"lifetime deal\" that expires when the vendor gets bored.
\n\n| App | Model | Price | First year | Three years |
|---|---|---|---|---|
| Jot | One-time licence | $49.99 | $49.99 | $49.99 |
| MacWhisper | One-time licence | €64 | €64 | €64 |
| Wispr Flow | Subscription | $12/mo, billed annually | $144 | $432 |
| Otter.ai | Subscription | $8.33/mo, billed annually | $99.99 | $299.97 |
| Fireflies.ai | Subscription | $10/mo, billed annually | $120 | $360 |
| tl;dv | Subscription | $18/mo, billed annually | $216 | $648 |
All prices are per user. Jot's licence is for one person on their own devices.
\n\n$49.99 divided by $12 a month is 4.2 months. Buy Jot and, by the fifth month of the subscription you did not buy, you have spent the same money and you do not own anything.
\nAgainst Otter's annual plan the break-even is about six months. Against tl;dv it is under three. There is no version of this arithmetic where the subscription wins on a multi-year horizon, because the subscription never stops.
\n\nThis is the part that makes the pricing question honest. A subscription is not a trick — it is what the architecture requires. If transcription runs on the vendor's servers, every minute of audio costs the vendor money in GPU time. There is no price at which they can sell you unlimited cloud inference forever, because their marginal cost is not zero.
\nA local model is different. Once the app is on your phone, the marginal cost of a transcription is a fraction of a battery charge. The vendor's cost is development, not inference — which is exactly the kind of cost a one-time price can fund.
\nSo the rule of thumb is simple and reliable:
\nWhen you see a cloud product offering a \"lifetime\" plan, ask what happens to inference costs in year three. The answer is usually a cap, a credit system, or a quiet discontinuation.
\n\nJot meets all five, which is a consequence of being local. MacWhisper meets them on macOS. Nothing else in the table does.
\n\n\n\nTeam administration. SSO, audit logs, a signed DPA, SOC 2 Type II, HIPAA. A support organisation with a phone number. If your procurement checklist contains those items, a one-time licence is not a shortcut around them — it is a product that cannot be bought at all, and you should buy the subscription and stop reading.
\nIt also does not buy the best accuracy on terrible audio. A frontier cloud model is better than a phone-sized model on overlapping speakers and heavy noise. A licence buys ownership, privacy, and a flat cost; it does not buy the best possible transcript in the worst possible conditions.
\n\nThe break-even formula is trivial: divide the licence price by the monthly subscription you would otherwise pay. If the answer is smaller than the number of months you expect to keep using the tool — and for a keyboard you use every day, it usually is — the licence wins on money before it wins on anything else.
\n\nTwo one-time options exist, and they are both local. That is not a coincidence; it is the economics of the category.
\nJot is $49.99 once and covers iPhone, iPad and iPod touch. MacWhisper is €64 once and covers macOS, which is the right answer if you want to batch-transcribe files on a Mac.
\nEverything else renews. If you are going to renew anyway, buy the one with the meeting bot you need — but know that you are paying for a server, and that the server is why.
\nIf you want the licence that runs on the phone in your pocket, Jot on the App Store.
\nEvery AI product you have used this year is priced the same way: per seat, per month. That is not a coincidence, and it is not only a business-model preference. It is what the cost structure looks like when every request runs on a GPU somebody else rents.
\nAn agent that transcribes your voice is a good example. A cloud model costs the vendor money for each minute of audio it processes. The subscription exists to cover that, plus the margin on top. Nothing sinister — just marginal cost.
\nNow move the model onto the phone. The marginal cost of a transcription becomes a fraction of a battery charge. There is no GPU bill to cover, no per-seat inference to amortise, and no reason the price has to recur.
\n\n| Cloud AI agent | On-device AI agent | |
|---|---|---|
| Where inference runs | Vendor's GPUs | Your phone |
| Vendor's cost per request | Real, and recurring | Effectively zero |
| Rational pricing | Subscription or metered credits | One-time licence |
| What you are buying | Continued access | The product |
| What happens if you stop paying | The product stops working | Nothing changes |
| What happens if the vendor folds | The product stops working | It keeps working |
That last row is the part people underestimate. A subscription is a lease on someone else's continuation. A licence to run a local model is a copy of the thing itself.
\n\nJot is an AI transcription agent that lives on the iPhone. It listens, transcribes on the device, and writes into whatever field you are in. The optional corrective pass is the only part that uses a network: a de-identified text snippet, no audio, sent to a US provider under a zero data retention agreement, and switchable off.
\nIt costs $49.99 once. There is also a $2.99 monthly plan for people who prefer that, which is worth mentioning because pretending a product only has one price is the kind of thing that erodes trust in a comparison.
\n\nLocal models are not as good as frontier models on hard audio. If your recordings are noisy, multi-speaker, or heavily accented, the cloud product will produce a better transcript, and no pricing model compensates for a transcript you have to correct by hand.
\nA one-time licence is also a bet on a vendor continuing to ship. That bet is smaller when the product runs locally — because a dead vendor leaves you with a working app — but it is not zero.
\n\nPer-seat AI pricing is defensible when the vendor pays for inference. It is much harder to defend when the model runs on your phone.
\nJot is the one-time version of a transcription agent: local model, local drafts, optional de-identified corrections, $49.99 once.
\nIf you want an AI tool you own rather than rent, Jot on the App Store.
\nThe product you know as Whisper Flow now resolves to Wispr Flow. whisperflow.com redirects to wisprflow.ai, and there is no separate Whisperflow product — anyone comparing against \"Whisperflow\" is comparing against this vendor under its old name.
\n\nBefore comparing any two tools it helps to say what is actually being claimed, because \"private\" gets used loosely enough to be meaningless. Three claims are separable, and they are separable on purpose:
\nOnly the third survives a determined attacker. The first two are policy commitments, and policy is exactly as strong as the vendor's controls on the day you need them. When a vendor writes a retention policy and a DPA, they are making a promise about a server they hold. When a product processes audio locally, there is no server, and the promise becomes a property of the architecture instead of a clause in a contract.
\nThis matters more than it sounds for the specific class of material that ends up in a transcription tool. Therapy sessions. Medical intake. Depositions. Job interviews. Board meetings. The audio is not incidental — the words are the sensitive part, and they are usually spoken by people who did not know a recorder was in the room.
\n\nWispr Flow's own documentation is unusually direct on this point, and I would rather quote it than paraphrase it around the edges:
\n\n“Transcription happens in the cloud rather than on your device, which is what enables real-time speed and accuracy.”
The architecture underneath that sentence is Cloud. Your microphone buffer is encoded and shipped to infrastructure the vendor operates, the model runs there, and the transcript comes back. That design buys real advantages — the frontier models available server-side are more accurate on hard audio than anything that fits in a laptop's memory budget, and there is no thermal throttling when a two-hour interview has to finish before you leave the building.
\nIt also means the material now exists somewhere that is not your laptop. Depending on the tier and the configuration, that can mean a retention window you set, a retention window the vendor set, or training on de-identified material by default.
\n\n| Question | Wispr Flow | Jot |
|---|---|---|
| Does audio leave the device? | Yes — cloud | No. Processing is local. |
| Is your content used for model training? | A setting you control; feedback pulls in session audio | Not applicable — nothing is transmitted. |
| Retention you control | Local history: keep, 24h or never — desktop only | Nothing retained, because nothing leaves. |
| Compliance posture | SOC 2 Type I complete; Type II audit complete, report pending | None claimed. See the honest note below. |
| Who holds a breach? | The vendor, and every subprocessor | You, and nobody else. |
Vendors use the phrase for at least three materially different positions, and the badges are not interchangeable.
\nThe transcription vendor keeps your data for the length of the job and deletes it immediately after, with a contractual zero-day policy flowing down the chain. This is a real and meaningful guarantee — but it only describes the seconds-to-minutes window inside a vendor you have already handed the audio to.
\nYour content is never used to improve a model. This is the single most load-bearing privacy clause in the category, and the strongest version names third parties explicitly, so the promise survives the vendor swapping model providers.
\nNothing is retained because nothing is transmitted. There is no copy on anyone's infrastructure, which means there is no retention window to configure, no DPA to negotiate, and no breach of yours to worry about on the vendor's side.
\nMost products in this comparison are at position 1 or 2. Position 3 is a different architectural claim, not a stronger version of the same one.
\n\nModel improvement is a separate switch from storage, feedback collects audio recordings and edits, and only an Enterprise contract turns data sharing off org-wide.
\nA genuinely configurable retention story for dictation — you can turn cloud storage off entirely, or set local deletion at 24 hours.
\nThat second point is worth stating plainly, because it is the reason Wispr Flow is still the right answer for a great many people. A vendor that publishes audit artefacts and honours a DPA is doing something a single-purpose product cannot. It is doing it for a different threat model — protecting you from a breach at a company that is contractually obliged to handle your data — and not for the threat model Jot is built against, which is the case where the company itself, or a compelled disclosure, or a future change of ownership, is the problem.
\n$280M Series B at a $2B valuation in August 2026, $361M raised in total — the largest balance sheet in the category. 270 of the Fortune 500; no public user count. Platform coverage is macOS, Windows, iOS, Android; Notetaker on Mac and Windows.
\nWorth naming plainly: a large, well-funded vendor has more resources to spend on security than a small one, and it is not honest to imply otherwise. The argument is not that Wispr Flow is careless. It is that a company whose business model is your ongoing subscription has a structural reason to keep your data in a useful place, and that reason does not evaporate when a policy page says otherwise.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nIf your recordings must not leave your device, the comparison is not close. Wispr Flow transmits your audio by design, and no configuration of that product changes the fact. There is no setting, no enterprise tier, and no DPA that converts a cloud architecture into a local one.
\nIf your recordings are ordinary — team meetings, project notes, nothing with a legal or clinical consequence — then Wispr Flow's certifications and integrations are worth their price, and you should buy it without reading this post.
\nThe decision is not about which vendor is more careful. It is about whether your threat model includes an adversary who has your data.
\nBefore comparing any two tools it helps to say what is actually being claimed, because \"private\" gets used loosely enough to be meaningless. Three claims are separable, and they are separable on purpose:
\nOnly the third survives a determined attacker. The first two are policy commitments, and policy is exactly as strong as the vendor's controls on the day you need them. When a vendor writes a retention policy and a DPA, they are making a promise about a server they hold. When a product processes audio locally, there is no server, and the promise becomes a property of the architecture instead of a clause in a contract.
\nThis matters more than it sounds for the specific class of material that ends up in a transcription tool. Therapy sessions. Medical intake. Depositions. Job interviews. Board meetings. The audio is not incidental — the words are the sensitive part, and they are usually spoken by people who did not know a recorder was in the room.
\n\nOtter.ai's own documentation is unusually direct on this point, and I would rather quote it than paraphrase it around the edges:
\n\n“We train our proprietary AI technology on de-identified audio recordings and on transcriptions.”
The architecture underneath that sentence is Cloud (AWS). Your microphone buffer is encoded and shipped to infrastructure the vendor operates, the model runs there, and the transcript comes back. That design buys real advantages — the frontier models available server-side are more accurate on hard audio than anything that fits in a laptop's memory budget, and there is no thermal throttling when a two-hour interview has to finish before you leave the building.
\nIt also means the material now exists somewhere that is not your laptop. Depending on the tier and the configuration, that can mean a retention window you set, a retention window the vendor set, or training on de-identified material by default.
\n\n| Question | Otter.ai | Jot |
|---|---|---|
| Does audio leave the device? | Yes — cloud (aws) | No. Processing is local. |
| Is your content used for model training? | On by default, automatic | Not applicable — nothing is transmitted. |
| Retention you control | No published figure; Trash clears at 30 days | Nothing retained, because nothing leaves. |
| Compliance posture | SOC 2 Type 2, GDPR, HIPAA on Enterprise | None claimed. See the honest note below. |
| Who holds a breach? | The vendor, and every subprocessor | You, and nobody else. |
Vendors use the phrase for at least three materially different positions, and the badges are not interchangeable.
\nThe transcription vendor keeps your data for the length of the job and deletes it immediately after, with a contractual zero-day policy flowing down the chain. This is a real and meaningful guarantee — but it only describes the seconds-to-minutes window inside a vendor you have already handed the audio to.
\nYour content is never used to improve a model. This is the single most load-bearing privacy clause in the category, and the strongest version names third parties explicitly, so the promise survives the vendor swapping model providers.
\nNothing is retained because nothing is transmitted. There is no copy on anyone's infrastructure, which means there is no retention window to configure, no DPA to negotiate, and no breach of yours to worry about on the vendor's side.
\nMost products in this comparison are at position 1 or 2. Position 3 is a different architectural claim, not a stronger version of the same one.
\n\nTraining on your audio is automatic and describes itself as such; no user-facing opt-out could be verified, and there is active litigation over exactly this.
\nThe deepest meeting-bot integrations and the broadest language support in the category, on the strength of a $100M ARR business.
\nThat second point is worth stating plainly, because it is the reason Otter.ai is still the right answer for a great many people. A vendor that publishes audit artefacts and honours a DPA is doing something a single-purpose product cannot. It is doing it for a different threat model — protecting you from a breach at a company that is contractually obliged to handle your data — and not for the threat model Jot is built against, which is the case where the company itself, or a compelled disclosure, or a future change of ownership, is the problem.
\n~$73M raised, last round February 2021, alongside a reported $100M ARR and 25M+ users. 25M+ users (vendor-reported). Platform coverage is Web, iOS, Android, Chrome extension, Zoom/Teams/Meet, MCP server, API.
\nWorth naming plainly: a large, well-funded vendor has more resources to spend on security than a small one, and it is not honest to imply otherwise. The argument is not that Otter.ai is careless. It is that a company whose business model is your ongoing subscription has a structural reason to keep your data in a useful place, and that reason does not evaporate when a policy page says otherwise.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nIf your recordings must not leave your device, the comparison is not close. Otter.ai transmits your audio by design, and no configuration of that product changes the fact. There is no setting, no enterprise tier, and no DPA that converts a cloud architecture into a local one.
\nIf your recordings are ordinary — team meetings, project notes, nothing with a legal or clinical consequence — then Otter.ai's certifications and integrations are worth their price, and you should buy it without reading this post.
\nThe decision is not about which vendor is more careful. It is about whether your threat model includes an adversary who has your data.
\nBefore comparing any two tools it helps to say what is actually being claimed, because \"private\" gets used loosely enough to be meaningless. Three claims are separable, and they are separable on purpose:
\nOnly the third survives a determined attacker. The first two are policy commitments, and policy is exactly as strong as the vendor's controls on the day you need them. When a vendor writes a retention policy and a DPA, they are making a promise about a server they hold. When a product processes audio locally, there is no server, and the promise becomes a property of the architecture instead of a clause in a contract.
\nThis matters more than it sounds for the specific class of material that ends up in a transcription tool. Therapy sessions. Medical intake. Depositions. Job interviews. Board meetings. The audio is not incidental — the words are the sensitive part, and they are usually spoken by people who did not know a recorder was in the room.
\n\nFireflies.ai's own documentation is unusually direct on this point, and I would rather quote it than paraphrase it around the edges:
\n\n“We never use your data for training.”
The architecture underneath that sentence is Cloud, transcription via Assembly.ai. Your microphone buffer is encoded and shipped to infrastructure the vendor operates, the model runs there, and the transcript comes back. That design buys real advantages — the frontier models available server-side are more accurate on hard audio than anything that fits in a laptop's memory budget, and there is no thermal throttling when a two-hour interview has to finish before you leave the building.
\nIt also means the material now exists somewhere that is not your laptop. Depending on the tier and the configuration, that can mean a retention window you set, a retention window the vendor set, or training on de-identified material by default.
\n\n| Question | Fireflies.ai | Jot |
|---|---|---|
| Does audio leave the device? | Yes — cloud, transcription via assembly.ai | No. Processing is local. |
| Is your content used for model training? | Never | Not applicable — nothing is transmitted. |
| Retention you control | Zero-day at subprocessors; your meetings are stored as the product | Nothing retained, because nothing leaves. |
| Compliance posture | SOC 2 Type II, ISO/IEC 42001:2023, GDPR, HIPAA, FERPA | None claimed. See the honest note below. |
| Who holds a breach? | The vendor, and every subprocessor | You, and nobody else. |
Vendors use the phrase for at least three materially different positions, and the badges are not interchangeable.
\nThe transcription vendor keeps your data for the length of the job and deletes it immediately after, with a contractual zero-day policy flowing down the chain. This is a real and meaningful guarantee — but it only describes the seconds-to-minutes window inside a vendor you have already handed the audio to.
\nYour content is never used to improve a model. This is the single most load-bearing privacy clause in the category, and the strongest version names third parties explicitly, so the promise survives the vendor swapping model providers.
\nNothing is retained because nothing is transmitted. There is no copy on anyone's infrastructure, which means there is no retention window to configure, no DPA to negotiate, and no breach of yours to worry about on the vendor's side.
\nMost products in this comparison are at position 1 or 2. Position 3 is a different architectural claim, not a stronger version of the same one.
\n\nEvery subprocessor is US-based with no EU data residency option, and custom retention is an Enterprise-only feature — so its zero-retention promise stops at the vendor boundary, not at the meeting.
\nThe most rigorous security posture in the category, and the only vendor here holding ISO/IEC 42001 for its AI management system. Its trust centre publishes the actual audit artefacts.
\nThat second point is worth stating plainly, because it is the reason Fireflies.ai is still the right answer for a great many people. A vendor that publishes audit artefacts and honours a DPA is doing something a single-purpose product cannot. It is doing it for a different threat model — protecting you from a breach at a company that is contractually obliged to handle your data — and not for the threat model Jot is built against, which is the case where the company itself, or a compelled disclosure, or a future change of ownership, is the problem.
\nOnly $19M ever raised, profitable since 2023, past a $1B valuation via a 2025 tender offer. 20M people, 500k+ organisations (vendor-reported). Platform coverage is Web, desktop (Mac and Windows), iOS, Android, Chrome extension, API, dialers.
\nWorth naming plainly: a large, well-funded vendor has more resources to spend on security than a small one, and it is not honest to imply otherwise. The argument is not that Fireflies.ai is careless. It is that a company whose business model is your ongoing subscription has a structural reason to keep your data in a useful place, and that reason does not evaporate when a policy page says otherwise.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nIf your recordings must not leave your device, the comparison is not close. Fireflies.ai transmits your audio by design, and no configuration of that product changes the fact. There is no setting, no enterprise tier, and no DPA that converts a cloud architecture into a local one.
\nIf your recordings are ordinary — team meetings, project notes, nothing with a legal or clinical consequence — then Fireflies.ai's certifications and integrations are worth their price, and you should buy it without reading this post.
\nThe decision is not about which vendor is more careful. It is about whether your threat model includes an adversary who has your data.
\nBefore comparing any two tools it helps to say what is actually being claimed, because \"private\" gets used loosely enough to be meaningless. Three claims are separable, and they are separable on purpose:
\nOnly the third survives a determined attacker. The first two are policy commitments, and policy is exactly as strong as the vendor's controls on the day you need them. When a vendor writes a retention policy and a DPA, they are making a promise about a server they hold. When a product processes audio locally, there is no server, and the promise becomes a property of the architecture instead of a clause in a contract.
\nThis matters more than it sounds for the specific class of material that ends up in a transcription tool. Therapy sessions. Medical intake. Depositions. Job interviews. Board meetings. The audio is not incidental — the words are the sensitive part, and they are usually spoken by people who did not know a recorder was in the room.
\n\ntl;dv's own documentation is unusually direct on this point, and I would rather quote it than paraphrase it around the edges:
\n\n“Tldx Solutions GmbH does not use Customer Content to train, fine-tune, or improve foundation models for the benefit of tldx Solutions GmbH or any third party.”
The architecture underneath that sentence is Cloud, EU data residency. Your microphone buffer is encoded and shipped to infrastructure the vendor operates, the model runs there, and the transcript comes back. That design buys real advantages — the frontier models available server-side are more accurate on hard audio than anything that fits in a laptop's memory budget, and there is no thermal throttling when a two-hour interview has to finish before you leave the building.
\nIt also means the material now exists somewhere that is not your laptop. Depending on the tier and the configuration, that can mean a retention window you set, a retention window the vendor set, or training on de-identified material by default.
\n\n| Question | tl;dv | Jot |
|---|---|---|
| Does audio leave the device? | Yes — cloud, eu data residency | No. Processing is local. |
| Is your content used for model training? | Never, contractually | Not applicable — nothing is transmitted. |
| Retention you control | Not published | Nothing retained, because nothing leaves. |
| Compliance posture | SOC 2 Type II, GDPR, EU AI Act; no HIPAA claim | None claimed. See the honest note below. |
| Who holds a breach? | The vendor, and every subprocessor | You, and nobody else. |
Vendors use the phrase for at least three materially different positions, and the badges are not interchangeable.
\nThe transcription vendor keeps your data for the length of the job and deletes it immediately after, with a contractual zero-day policy flowing down the chain. This is a real and meaningful guarantee — but it only describes the seconds-to-minutes window inside a vendor you have already handed the audio to.
\nYour content is never used to improve a model. This is the single most load-bearing privacy clause in the category, and the strongest version names third parties explicitly, so the promise survives the vendor swapping model providers.
\nNothing is retained because nothing is transmitted. There is no copy on anyone's infrastructure, which means there is no retention window to configure, no DPA to negotiate, and no breach of yours to worry about on the vendor's side.
\nMost products in this comparison are at position 1 or 2. Position 3 is a different architectural claim, not a stronger version of the same one.
\n\nAll of that protection is defeated in practice by an August 2026 exposure of roughly 181,874 meeting records through a missing database authorisation rule. Residency and contracts do not substitute for correct access control.
\nThe cleanest no-training clause in the category — it extends the prohibition to any third party, not just to tl;dv's own models — plus genuine EU data residency for buyers who need it.
\nThat second point is worth stating plainly, because it is the reason tl;dv is still the right answer for a great many people. A vendor that publishes audit artefacts and honours a DPA is doing something a single-purpose product cannot. It is doing it for a different threat model — protecting you from a breach at a company that is contractually obliged to handle your data — and not for the threat model Jot is built against, which is the case where the company itself, or a compelled disclosure, or a future change of ownership, is the problem.
\nA €4.3M seed round in 2022 and 3M+ users — by far the thinnest capitalisation of the five, against the largest user-facing trust surface. 3M+ users (vendor-reported). Platform coverage is Web, desktop, iOS, Android, Chrome extension, Zoom/Meet/Teams, API and webhooks, MCP server.
\nWorth naming plainly: a large, well-funded vendor has more resources to spend on security than a small one, and it is not honest to imply otherwise. The argument is not that tl;dv is careless. It is that a company whose business model is your ongoing subscription has a structural reason to keep your data in a useful place, and that reason does not evaporate when a policy page says otherwise.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nIf your recordings must not leave your device, the comparison is not close. tl;dv transmits your audio by design, and no configuration of that product changes the fact. There is no setting, no enterprise tier, and no DPA that converts a cloud architecture into a local one.
\nIf your recordings are ordinary — team meetings, project notes, nothing with a legal or clinical consequence — then tl;dv's certifications and integrations are worth their price, and you should buy it without reading this post.
\nThe decision is not about which vendor is more careful. It is about whether your threat model includes an adversary who has your data.
\nBefore comparing any two tools it helps to say what is actually being claimed, because \"private\" gets used loosely enough to be meaningless. Three claims are separable, and they are separable on purpose:
\nOnly the third survives a determined attacker. The first two are policy commitments, and policy is exactly as strong as the vendor's controls on the day you need them. When a vendor writes a retention policy and a DPA, they are making a promise about a server they hold. When a product processes audio locally, there is no server, and the promise becomes a property of the architecture instead of a clause in a contract.
\nThis matters more than it sounds for the specific class of material that ends up in a transcription tool. Therapy sessions. Medical intake. Depositions. Job interviews. Board meetings. The audio is not incidental — the words are the sensitive part, and they are usually spoken by people who did not know a recorder was in the room.
\n\nMacWhisper's own documentation is unusually direct on this point, and I would rather quote it than paraphrase it around the edges:
\n\n“MacWhisper does all its functionality on your device. No data — audio, text or other — leaves your device.”
The architecture underneath that sentence is On-device, local models. Your microphone buffer is encoded and shipped to infrastructure the vendor operates, the model runs there, and the transcript comes back. That design buys real advantages — the frontier models available server-side are more accurate on hard audio than anything that fits in a laptop's memory budget, and there is no thermal throttling when a two-hour interview has to finish before you leave the building.
\nIt also means the material now exists somewhere that is not your laptop. Depending on the tier and the configuration, that can mean a retention window you set, a retention window the vendor set, or training on de-identified material by default.
\n\n| Question | MacWhisper | Jot |
|---|---|---|
| Does audio leave the device? | Yes — on-device, local models | No. Processing is local. |
| Is your content used for model training? | Not applicable, nothing leaves | Not applicable — nothing is transmitted. |
| Retention you control | Nothing retained, because nothing is transmitted | Nothing retained, because nothing leaves. |
| Compliance posture | None — no SOC 2, GDPR, HIPAA, DPA or trust centre | None claimed. See the honest note below. |
| Who holds a breach? | The vendor, and every subprocessor | You, and nobody else. |
Vendors use the phrase for at least three materially different positions, and the badges are not interchangeable.
\nThe transcription vendor keeps your data for the length of the job and deletes it immediately after, with a contractual zero-day policy flowing down the chain. This is a real and meaningful guarantee — but it only describes the seconds-to-minutes window inside a vendor you have already handed the audio to.
\nYour content is never used to improve a model. This is the single most load-bearing privacy clause in the category, and the strongest version names third parties explicitly, so the promise survives the vendor swapping model providers.
\nNothing is retained because nothing is transmitted. There is no copy on anyone's infrastructure, which means there is no retention window to configure, no DPA to negotiate, and no breach of yours to worry about on the vendor's side.
\nMost products in this comparison are at position 1 or 2. Position 3 is a different architectural claim, not a stronger version of the same one.
\n\nNo compliance certifications at all, which rules it out for regulated procurement outright, and macOS is a hard ceiling with no Windows, Linux, or web build.
\nIt is the only product here that is genuinely local, and the only one that genuinely sells a licence rather than a subscription. On the two axes this comparison measures, it is the harder competitor.
\nThat second point is worth stating plainly, because it is the reason MacWhisper is still the right answer for a great many people. A vendor that publishes audit artefacts and honours a DPA is doing something a single-purpose product cannot. It is doing it for a different threat model — protecting you from a breach at a company that is contractually obliged to handle your data — and not for the threat model Jot is built against, which is the case where the company itself, or a compelled disclosure, or a future change of ownership, is the problem.
\nAn independent, bootstrapped product from Good Snooze in the Netherlands, with no institutional funding behind it and no trust centre to publish into. Featured in Apple's Mac mini newsroom; no public user count. Platform coverage is macOS only, plus a separate iOS and iPad App Store build.
\nWorth naming plainly: a large, well-funded vendor has more resources to spend on security than a small one, and it is not honest to imply otherwise. The argument is not that MacWhisper is careless. It is that a company whose business model is your ongoing subscription has a structural reason to keep your data in a useful place, and that reason does not evaporate when a policy page says otherwise.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nIf your recordings must not leave your device, the comparison is not close. MacWhisper transmits your audio by design, and no configuration of that product changes the fact. There is no setting, no enterprise tier, and no DPA that converts a cloud architecture into a local one.
\nIf your recordings are ordinary — team meetings, project notes, nothing with a legal or clinical consequence — then MacWhisper's certifications and integrations are worth their price, and you should buy it without reading this post.
\nThe decision is not about which vendor is more careful. It is about whether your threat model includes an adversary who has your data.
\nThe product you know as Whisper Flow now resolves to Wispr Flow. whisperflow.com redirects to wisprflow.ai, and there is no separate Whisperflow product — anyone comparing against \"Whisperflow\" is comparing against this vendor under its old name.
\n\nWispr Flow is a good product. This is not a takedown — it is a calculation, and the calculation is unfavourable in a specific and quantifiable way that is worth understanding before you sign up for something that renews automatically.
\n\nEvery figure below is the vendor's own published price, read on 2026-10-18, with the annual plan shown at its effective monthly rate — the number you are actually billed per month once the discount is applied, not the headline monthly figure. Prices in this category move often, so each post links the live pricing page rather than treating any number here as permanent.
\nThree comparisons matter, and only the first gets discussed publicly:
\nNote also that the one-time price is not risk-free in the buyer's favour: it trades a known cost for an unknown future one. What that trade is actually worth is a separate argument, and the honest version of it is made in the verdict rather than hidden.
\n\n| Jot | Wispr Flow | |
|---|---|---|
| One-time purchase | $49.99 | None — no licence is sold |
| Monthly, billed monthly | $2.99 | $15 per user |
| Monthly, billed annually | — | $12 per user |
| Annual total | $49.99 | $144 |
| Free tier | The product is the product | The free tier is not a trial that expires — it is capped by volume at 2,000 words per week on desktop and 1,000 per week on mobile, with Notetaker access limited. |
| After three years | $49.99 | $432 |
| Cost per year over three years | $16.66 | $12.00 |
$49.99 divided by $12 per month is 4.2 months. Buy Jot at $49.99, and by somewhere around the 5th month of using Wispr Flow instead, you will have spent the same money. Every month after that is money you already paid for and are spending again.
\nThe awkward way to put this is that Wispr Flow is cheaper than Jot for the first 4 months of use, and more expensive for every month after. So the question is not whether the subscription is affordable. It is whether you will still be using it in 5 months — and for a tool people keep for years, that is not a hard question.
\nJot is also available at $2.99 per month, which is worth being straight about. It is there for people who genuinely prefer to spread the cost, and it is competitive — $2.99 sits below every paid tier in this comparison except Otter's student rate. But it is worth noticing that even this option loses to the one-time price after about 17 months: three years of the monthly plan costs $108, against $49.99 paid once.
\nThat gap is the entire argument for owning software rather than renting it, expressed in one number.
\n\nThe table above compares like with like: one seat, one person, the features you would actually use. A fair comparison has to say what it leaves out.
\nNone of these are reasons to dismiss the comparison. They are the conditions under which it holds. If you are one person, or a very small team, on a single device, and you will keep using the tool past month 5, the maths runs one way and only one way.
\n\nA subscription is a relationship, not a transaction, and the difference shows up in places a spreadsheet cannot see.
\nNone of this makes a subscription illegitimate. Plenty of teams should absolutely buy Otter or Fireflies and should be glad the vendor is solvent. The argument is only that a subscription is not the only way to own software, and that a tool which never phones home has no reason to charge you rent for permission to keep existing.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nWispr Flow costs less than Jot for the first 4 months and more than Jot from month 5 onward. Over three years of ordinary use, $432 against $49.99 is roughly 8.6× the cost for a product that does the same job locally.
\nWhat you give up: macOS, Windows, iOS, Android; Notetaker on Mac and Windows, plus the team and compliance infrastructure a subscription-funded vendor can afford to maintain. On a regulated procurement checklist, Wispr Flow may be the only purchasable option, and no arithmetic changes that.
\nWhat you get back: a tool you own, that keeps working if the vendor does not, that never sends your audio to anyone, and that you stop paying for in year two.
\nOtter.ai is a good product. This is not a takedown — it is a calculation, and the calculation is unfavourable in a specific and quantifiable way that is worth understanding before you sign up for something that renews automatically.
\n\nEvery figure below is the vendor's own published price, read on 2026-10-18, with the annual plan shown at its effective monthly rate — the number you are actually billed per month once the discount is applied, not the headline monthly figure. Prices in this category move often, so each post links the live pricing page rather than treating any number here as permanent.
\nThree comparisons matter, and only the first gets discussed publicly:
\nNote also that the one-time price is not risk-free in the buyer's favour: it trades a known cost for an unknown future one. What that trade is actually worth is a separate argument, and the honest version of it is made in the verdict rather than hidden.
\n\n| Jot | Otter.ai | |
|---|---|---|
| One-time purchase | $49.99 | None — no licence is sold |
| Monthly, billed monthly | $2.99 | $16.99 per user |
| Monthly, billed annually | — | $8.33 per user |
| Annual total | $49.99 | $99.99 |
| Free tier | The product is the product | Otter Basic is free forever but throttled hard: 300 transcription minutes a month, and three file imports for the entire life of the account. |
| After three years | $49.99 | $300 |
| Cost per year over three years | $16.66 | $8.33 |
$49.99 divided by $8.33 per month is 6.0 months. Buy Jot at $49.99, and by somewhere around the 7th month of using Otter.ai instead, you will have spent the same money. Every month after that is money you already paid for and are spending again.
\nThe awkward way to put this is that Otter.ai is cheaper than Jot for the first 6 months of use, and more expensive for every month after. So the question is not whether the subscription is affordable. It is whether you will still be using it in 7 months — and for a tool people keep for years, that is not a hard question.
\nJot is also available at $2.99 per month, which is worth being straight about. It is there for people who genuinely prefer to spread the cost, and it is competitive — $2.99 sits below every paid tier in this comparison except Otter's student rate. But it is worth noticing that even this option loses to the one-time price after about 17 months: three years of the monthly plan costs $108, against $49.99 paid once.
\nThat gap is the entire argument for owning software rather than renting it, expressed in one number.
\n\nThe table above compares like with like: one seat, one person, the features you would actually use. A fair comparison has to say what it leaves out.
\nNone of these are reasons to dismiss the comparison. They are the conditions under which it holds. If you are one person, or a very small team, on a single device, and you will keep using the tool past month 7, the maths runs one way and only one way.
\n\nA subscription is a relationship, not a transaction, and the difference shows up in places a spreadsheet cannot see.
\nNone of this makes a subscription illegitimate. Plenty of teams should absolutely buy Otter or Fireflies and should be glad the vendor is solvent. The argument is only that a subscription is not the only way to own software, and that a tool which never phones home has no reason to charge you rent for permission to keep existing.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nOtter.ai costs less than Jot for the first 6 months and more than Jot from month 7 onward. Over three years of ordinary use, $300 against $49.99 is roughly 6.0× the cost for a product that does the same job locally.
\nWhat you give up: Web, iOS, Android, Chrome extension, Zoom/Teams/Meet, MCP server, API, plus the team and compliance infrastructure a subscription-funded vendor can afford to maintain. On a regulated procurement checklist, Otter.ai may be the only purchasable option, and no arithmetic changes that.
\nWhat you get back: a tool you own, that keeps working if the vendor does not, that never sends your audio to anyone, and that you stop paying for in year two.
\nFireflies.ai is a good product. This is not a takedown — it is a calculation, and the calculation is unfavourable in a specific and quantifiable way that is worth understanding before you sign up for something that renews automatically.
\n\nEvery figure below is the vendor's own published price, read on 2026-10-18, with the annual plan shown at its effective monthly rate — the number you are actually billed per month once the discount is applied, not the headline monthly figure. Prices in this category move often, so each post links the live pricing page rather than treating any number here as permanent.
\nThree comparisons matter, and only the first gets discussed publicly:
\nNote also that the one-time price is not risk-free in the buyer's favour: it trades a known cost for an unknown future one. What that trade is actually worth is a separate argument, and the honest version of it is made in the verdict rather than hidden.
\n\n| Jot | Fireflies.ai | |
|---|---|---|
| One-time purchase | $49.99 | None — no licence is sold |
| Monthly, billed monthly | $2.99 | $18 per user |
| Monthly, billed annually | — | $10 per user |
| Annual total | $49.99 | $120 |
| Free tier | The product is the product | The free tier records and transcribes without a hard minute cap, but storage is 400 minutes for the whole team — so it is free right up until you want to keep anything. |
| After three years | $49.99 | $360 |
| Cost per year over three years | $16.66 | $10.00 |
$49.99 divided by $10 per month is 5.0 months. Buy Jot at $49.99, and by somewhere around the 5th month of using Fireflies.ai instead, you will have spent the same money. Every month after that is money you already paid for and are spending again.
\nThe awkward way to put this is that Fireflies.ai is cheaper than Jot for the first 4 months of use, and more expensive for every month after. So the question is not whether the subscription is affordable. It is whether you will still be using it in 5 months — and for a tool people keep for years, that is not a hard question.
\nJot is also available at $2.99 per month, which is worth being straight about. It is there for people who genuinely prefer to spread the cost, and it is competitive — $2.99 sits below every paid tier in this comparison except Otter's student rate. But it is worth noticing that even this option loses to the one-time price after about 17 months: three years of the monthly plan costs $108, against $49.99 paid once.
\nThat gap is the entire argument for owning software rather than renting it, expressed in one number.
\n\nThe table above compares like with like: one seat, one person, the features you would actually use. A fair comparison has to say what it leaves out.
\nNone of these are reasons to dismiss the comparison. They are the conditions under which it holds. If you are one person, or a very small team, on a single device, and you will keep using the tool past month 5, the maths runs one way and only one way.
\n\nA subscription is a relationship, not a transaction, and the difference shows up in places a spreadsheet cannot see.
\nNone of this makes a subscription illegitimate. Plenty of teams should absolutely buy Otter or Fireflies and should be glad the vendor is solvent. The argument is only that a subscription is not the only way to own software, and that a tool which never phones home has no reason to charge you rent for permission to keep existing.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nFireflies.ai costs less than Jot for the first 4 months and more than Jot from month 5 onward. Over three years of ordinary use, $360 against $49.99 is roughly 7.2× the cost for a product that does the same job locally.
\nWhat you give up: Web, desktop (Mac and Windows), iOS, Android, Chrome extension, API, dialers, plus the team and compliance infrastructure a subscription-funded vendor can afford to maintain. On a regulated procurement checklist, Fireflies.ai may be the only purchasable option, and no arithmetic changes that.
\nWhat you get back: a tool you own, that keeps working if the vendor does not, that never sends your audio to anyone, and that you stop paying for in year two.
\ntl;dv is a good product. This is not a takedown — it is a calculation, and the calculation is unfavourable in a specific and quantifiable way that is worth understanding before you sign up for something that renews automatically.
\n\nEvery figure below is the vendor's own published price, read on 2026-10-18, with the annual plan shown at its effective monthly rate — the number you are actually billed per month once the discount is applied, not the headline monthly figure. Prices in this category move often, so each post links the live pricing page rather than treating any number here as permanent.
\nThree comparisons matter, and only the first gets discussed publicly:
\nNote also that the one-time price is not risk-free in the buyer's favour: it trades a known cost for an unknown future one. What that trade is actually worth is a separate argument, and the honest version of it is made in the verdict rather than hidden.
\n\n| Jot | tl;dv | |
|---|---|---|
| One-time purchase | $49.99 | None — no licence is sold |
| Monthly, billed monthly | $2.99 | $29 per user |
| Monthly, billed annually | — | $18 per user |
| Annual total | $49.99 | $216 |
| Free tier | The product is the product | The free tier is the most generous on paper in this comparison: unlimited recordings and unlimited transcription across 30+ languages, with AI notes capped at ten meetings a month. |
| After three years | $49.99 | $648 |
| Cost per year over three years | $16.66 | $18.00 |
$49.99 divided by $18 per month is 2.8 months. Buy Jot at $49.99, and by somewhere around the 3rd month of using tl;dv instead, you will have spent the same money. Every month after that is money you already paid for and are spending again.
\nThe awkward way to put this is that tl;dv is cheaper than Jot for the first 2 months of use, and more expensive for every month after. So the question is not whether the subscription is affordable. It is whether you will still be using it in 3 months — and for a tool people keep for years, that is not a hard question.
\nJot is also available at $2.99 per month, which is worth being straight about. It is there for people who genuinely prefer to spread the cost, and it is competitive — $2.99 sits below every paid tier in this comparison except Otter's student rate. But it is worth noticing that even this option loses to the one-time price after about 17 months: three years of the monthly plan costs $108, against $49.99 paid once.
\nThat gap is the entire argument for owning software rather than renting it, expressed in one number.
\n\nThe table above compares like with like: one seat, one person, the features you would actually use. A fair comparison has to say what it leaves out.
\nNone of these are reasons to dismiss the comparison. They are the conditions under which it holds. If you are one person, or a very small team, on a single device, and you will keep using the tool past month 3, the maths runs one way and only one way.
\n\nA subscription is a relationship, not a transaction, and the difference shows up in places a spreadsheet cannot see.
\nNone of this makes a subscription illegitimate. Plenty of teams should absolutely buy Otter or Fireflies and should be glad the vendor is solvent. The argument is only that a subscription is not the only way to own software, and that a tool which never phones home has no reason to charge you rent for permission to keep existing.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\ntl;dv costs less than Jot for the first 2 months and more than Jot from month 3 onward. Over three years of ordinary use, $648 against $49.99 is roughly 13.0× the cost for a product that does the same job locally.
\nWhat you give up: Web, desktop, iOS, Android, Chrome extension, Zoom/Meet/Teams, API and webhooks, MCP server, plus the team and compliance infrastructure a subscription-funded vendor can afford to maintain. On a regulated procurement checklist, tl;dv may be the only purchasable option, and no arithmetic changes that.
\nWhat you get back: a tool you own, that keeps working if the vendor does not, that never sends your audio to anyone, and that you stop paying for in year two.
\nThis is the awkward one to write, and I would rather be direct about why than dress it up.
\nAlmost every other comparison in this series is structurally easy: a subscription vendor on one side, a one-time licence on the other, and a break-even month that does the argument for you. MacWhisper is already a one-time purchase. There is no subscription to break even from. The easy version of this post would quietly invent one — quote the per-user seat price of some other tool, imply MacWhisper rents you access, and produce a tidy conclusion. That would be false, and you would eventually find out.
\n\n| Jot | MacWhisper | |
|---|---|---|
| One-time purchase | $49.99 | €64 per licence |
| Recurring cost | None | None |
| Updates | Included | Lifetime updates included |
| Free version | — | Full app, capped model size, free forever |
| Volume pricing | — | 5 licences €54 · 10 €49 · 20 €45 · 50 €44 |
| Platform | See the product page | macOS only, plus iOS/iPad |
| Compliance certifications | None claimed | None claimed |
The dollar figure shown by MacWhisper's checkout varies by region — €64 and roughly $73 are the same product localised differently, not two prices. At any exchange rate in the plausible range, MacWhisper is more expensive than Jot for a single user.
\n\nBoth products process audio locally. This is the part of the comparison that is genuinely close: MacWhisper's privacy policy states that no data leaves your device, and Jot makes the same architectural claim. Neither has a compliance certification. On the axis that matters most here, the $14 or so you save with Jot buys you nothing.
\nIf you are choosing on privacy alone and you own a Mac, MacWhisper is a legitimate choice and this post is not going to pretend otherwise.
\nMacWhisper runs Whisper, Qwen, and NVIDIA Parakeet locally, uses NVIDIA Nemotron for diarisation, and reaches roughly 30× realtime on Apple Silicon. It also lets you bring your own key for OpenAI, Anthropic, Deepgram, ElevenLabs, and about a dozen more, or point it at Ollama and LM Studio for a fully local large-model setup. That is a mature and well-considered model-selection story.
\nA subscription is a relationship, not a transaction, and the difference shows up in places a spreadsheet cannot see.
\nNone of this makes a subscription illegitimate. Plenty of teams should absolutely buy Otter or Fireflies and should be glad the vendor is solvent. The argument is only that a subscription is not the only way to own software, and that a tool which never phones home has no reason to charge you rent for permission to keep existing.
\n\nIt is worth being explicit about what has been left out. Four of the five posts in this series argue that a one-time licence breaks even against a subscription within a few months. That argument does not apply here, because there is no subscription on the other side of the table. Applying it anyway would inflate the case and damage it — a reader who catches one fabricated comparison stops trusting all ten.
\nWhat applies instead is simpler and less exciting: two local products, comparable privacy architectures, similar absence of certifications, and a price difference of roughly $14.00 on a single purchase. Over a year of ownership that difference approaches zero. Judge them on the product.
\n\nA comparison written only by the winner is marketing, and it should not be trusted. Here is the case against Jot, stated as well as its proponents would state it.
\nThe vendor under comparison here has engineers, a support organisation, a security team, and a compliance budget. Jot is a single-privacy-focused product with a price tag and a promise. When your procurement checklist demands SOC 2 Type II, ISO 27001, a signed DPA, a HIPAA BAA, a SAML SSO integration, and an audit log, the incumbent is not merely easier to buy — it may be the only thing you are permitted to buy.
\nTeam features are the same story. Shared workspaces, admin controls, seat management, centralised billing, and an audit trail are real engineering, and Jot does not attempt them. If you are coordinating a department, an incumbent tool is the correct answer.
\nModel quality on difficult audio — overlapping speakers, heavy accents, bad microphones, background noise — still favours large frontier models that run on server GPUs. On-device models have narrowed the gap substantially, and they have not closed it.
\nAll of that is true, and none of it is an argument against Jot for the person whose recordings must not leave their laptop. It is an argument about who should buy what. My position is that the two questions — where does the audio go and who has to be able to buy this — have different answers for different people, and that conflating them is how a good privacy posture ends up described as insufficient for work.
\n\nThis is the comparison where I have the least to offer you. MacWhisper is genuinely local, genuinely a one-time purchase, and genuinely mature. Jot does not beat it on architecture and costs about $14.00 less.
\nWhere Jot earns the difference: if you are not on macOS, MacWhisper is not an option. If you want an agent rather than a transcription tool, the shapes differ. And if you already use one of the other four products in this series and know the subscription arithmetic, you have probably worked out where you land.
\nWhere MacWhisper wins outright: longevity, platform maturity, model library breadth, and the fact that it has been doing this for years. If you are choosing one local transcription tool and you own a Mac, it is the harder recommendation to argue against.
\nPrivacy pages are written to be read generously. This is the opposite: a list of questions with answers you can verify without trusting anyone's copywriter. Work through them in order. The first one disqualifies most of the category.
\n\nHow to check: turn on airplane mode, dictate a sentence.
\nIf text appears, the model is on the device. If nothing appears, the audio is being uploaded, and every subsequent question is about how responsibly that upload is handled rather than whether it happens.
\n\nHow to check: search the privacy policy and docs for \"cloud\", \"on-device\", and \"local\".
\nWispr Flow is admirably direct: “Transcription always occurs on the cloud.” MacWhisper is equally direct in the other direction. A vendor that will not say is telling you something.
\n\nHow to check: look for the words \"train\", \"improve\", \"model improvement\", and \"feedback\" in the policy.
\nWispr Flow's Data Controls page is explicit that feedback collects \"audio recordings, transcripts, your edits\" and that this may be used to \"develop and train our AI models\". It also states plainly that storage is \"controlled by the applicable Service features, not by your model training preference\" — two switches, not one.
\n\nHow to check: read the feedback clause, not the settings page.
\nIn a cloud product, a rating is often a data-sharing action that attaches the session. That is fine if you know; it is not fine if you think you are filing a bug report.
\n\nHow to check: find a retention section with a number in it.
\n\"Configurable\" is better than nothing. \"Generally 30 days\" is not a number you can hold anyone to. Wispr Flow offers three local-history options on desktop — keep, 24 hours, never — and describes them as \"desktop today, with mobile coming soon\".
\n\nHow to check: look for a per-platform note. Retention settings frequently ship on desktop first.
\nThis is the single most common gap between what a vendor says and what a phone user gets.
\n\nHow to check: read the sentence around the acronym and look for exclusions.
\nWispr Flow says it maintains ZDR agreements with third-party AI providers — and that \"certain features, such as briefs for Notetaker\" are not supported under ZDR. That is a subcontractor guarantee with a published carve-out, not a property of the product.
\n\nHow to check: search the policy for \"context\", \"active window\", or \"app\".
\nWispr Flow discloses both: optional Context Awareness may read \"the text on the screen\", and separately the app \"always uses information about the app you are dictating in\" plus \"the surrounding textbox content\".
\n\nHow to check: ask for the SOC 2 report, the DPA, and the subprocessor list. Not the badge — the document.
\nFireflies publishes audit artefacts through its trust centre. Wispr Flow sells Type II and ISO 27001 at the Enterprise tier. Jot has none of this and does not pretend otherwise, which rules it out for regulated procurement and is worth saying plainly.
\n\nHow to check: cancel the plan in your head and reread the export and deletion sections.
\nA subscription answers this with a vendor policy. A local app answers it with a folder on your disk.
\n\n| App | Local model | Training | Local retention control | Compliance artefact | Price |
|---|---|---|---|---|---|
| Jot | Yes | Not applicable | On-device by default | None claimed | $49.99 once |
| Wispr Flow | No — cloud | Setting, plus feedback path | Desktop only, \"mobile coming soon\" | SOC 2 Type II at Enterprise | $144 a year |
| Otter.ai | No — cloud | On by default | Not published | SOC 2 Type 2, HIPAA on Enterprise | $99.99 a year |
| Fireflies.ai | No — cloud | Never, per policy | Enterprise-only custom retention | SOC 2 Type II, ISO 42001 | $120 a year |
| tl;dv | No — cloud | Never, contractually | Not published | SOC 2 Type II, EU AI Act | $216 a year |
| MacWhisper | Yes, macOS | Not applicable | On-device by default | None claimed | €64 once |
Questions 1 through 4 decide whether the recording is private in the sense most people mean. Questions 5 through 8 decide whether the vendor's promises are narrow or broad. Questions 9 and 10 decide whether the tool is purchasable at all in an organisation.
\nFor a personal dictation keyboard, the first group is the whole decision and the rest is detail. For a regulated buyer, the last group overrides everything, and no amount of local processing substitutes for a signed DPA.
\n\nA vendor that answers all ten is rare. A vendor whose answers do not depend on a policy page is rarer.
\nIf you want the answers that cannot change, they are the architectural ones: the model on your device, the transcript on your disk, the audio never uploaded.
\nThat is Jot on the App Store.
\n