Whisperstream, and then some
Whisperstream is the comparison we like best, because it gets the important thing right. Its transcription runs on your CPU, its privacy policy says your microphone audio "is never uploaded to any server", there is no account, and it says "No telemetry, no analytics, error reporting, crash reporting, no usage tracking". If you are choosing between it and a cloud dictation service, pick Whisperstream. This page is about the narrower question: Whisperstream or Deixis, both local, both on Windows, and Deixis on the Mac and your phone as well.
The short version: Deixis is free for everything that runs on your PC, where Whisperstream is $29 after a seven-day trial. Deixis knows which app and which field you are dictating into and punctuates accordingly, with nothing to set up. Its vocabulary steers recognition itself rather than correcting the text afterwards. And it is more than a dictation box: the same key records meetings from both sides of a call, grabs and annotates screenshots, and records the screen, all on your PC.
Whisperstream wins some of this back, and the list is below: 39 languages to Deixis's English, a local cleanup model it downloads for you, cleanup prompts of your own, an encrypted history with a retention window. If you dictate in Spanish, stop reading and buy Whisperstream. Otherwise, read on.
Why switch
What you gain, and what you give up
Why Deixis
- Free, for good. Everything on your PC is free with no trial clock and no account: dictation, hands-free notes, meetings, screenshots, screen recording, the dictionary and all five vocabulary packs. Whisperstream's trial is seven days from your first successful dictation, then $29 for one device or $49 for two, and until you pay, recording stops. Deixis Pro, at $49 a year or $10 a month, is only for the managed cloud extras, and even those are free with your own provider key.
- One tool, on every device you own. Windows, Mac, iPhone and Android, with the same on-device speech recognition on all four; your audio never leaves any of them. On the iPhone it is a keyboard with a microphone key, on Android a floating microphone that dictates into any app. Whisperstream is a Windows program.
- It knows which field you are in, with nothing to set up. A commit message, a chat reply and a ticket are punctuated differently, and Deixis does that on its own because it knows which app and which control the text is going into. Only the names of the app and the field ride along with a cleanup request, never the window title and never a screenshot. Whisperstream can route a cleanup profile by application, but you write the prompt and the rule, its enhancement is off until you switch it on, and it sees the app, not the field.
- Your vocabulary steers recognition, not just the text after it. Whisperstream's dictionary is
filters and replacements applied to completed dictation, and a new install starts with it empty.
Deixis feeds your terms into the speech model's prompt, so a product name or a library is heard
correctly in the first place, keeps replacements for the rest, and ships five vocabulary packs,
for software, finance, business, medicine and law. Select a word anywhere, press
Ctrl+Win+D, done. - Meetings from both sides of the call.
Ctrl+Win+Mrecords your microphone and what your speakers play, transcribes both on your PC, and writes a timestamped transcript to a folder. The remote voices are told apart on the device; name one and it is recognised in later meetings, from a voiceprint that stays on your machine, is never uploaded and can be deleted per person. No bot joins, nothing uploads. Whisperstream does not record meetings. - Screenshots, screen recording, and talking while you point.
Ctrl+Win+Sfreezes the screen, drag a region, add arrows, boxes, numbered steps or a pixelated redaction, and save a PNG that stays re-editable.Ctrl+Win+Vrecords the screen to a video, and if a microphone is set for recordings it transcribes what you said over it; without one the recording is silent. Free mode lets you narrate for as long as you like while grabbing a region, picking an element on screen or taking your selection, and hands you one note that interleaves what you said with what you pointed at. Whisperstream does none of this. - Buttons and webhooks. Custom buttons on the palette and toolbar run a script of yours or post to a URL you own, and can type the result back at the caret; chains string steps together. Whisperstream's workflows are built in, a pin mode and spoken replies for coding-agent sessions; nothing in its docs runs a program of yours or posts to a URL.
- No licence check. Whisperstream's privacy policy says a Pro licence sends the key, hostname and OS family for validation about once a week while online. Deixis has no licence to validate for anything local; the only automatic call is an update check that sends nothing but the request.
Where Whisperstream still wins
- Thirty-nine languages. Whisperstream's docs list 39 dictation languages across its Parakeet and Qwen3 speech models, and the multilingual models detect the language per recording. Deixis's speech models are English models.
- A local cleanup model in the box. Whisperstream's Local enhancement provider downloads a small language model, Gemma, and runs it on your PC with no key and no server to set up, after a speed check that says whether your machine is up to it. Deixis's local cleanup runs as rules; a language model on your own machine means pointing Deixis at a model server you run yourself.
- Cleanup prompts you write yourself. Enhancement profiles carry your own cleanup instructions, or a chat, code or email template as a start, and rules route them by the website or the application in front. Deixis has one cleanup behaviour and no prompt to edit; it reads the app and field for you instead.
- More cloud providers, under your key. Gemini, OpenAI, Anthropic, Ollama and a custom endpoint. Deixis keeps it to one shape: Deixis's cloud, your own key through the same route, or a local server.
- An encrypted history with a retention window. Dictation history is encrypted at rest on your device and kept for 1, 7 or 30 days, or forever, with 7 days as the default. Deixis keeps a history file you can read and export, without a retention timer.
- Pay once. $29 one-time for one device, $49 for two, with a 30-day money-back guarantee. Deixis is free for everything local and charges a subscription only for the managed cloud extras.
- Two ways to insert text. Whisperstream offers clipboard paste or simulated keystrokes, one character at a time, for screen readers, remote desktops and places where the clipboard is blocked. Deixis pastes at the caret and restores your clipboard.
What you get
More than dictation. One key, and everything behind it.
Hold the key and press a letter: free mode, a grab, a meeting, a recording, a translation, your own buttons. All of it on your own device, all of it free, none of it behind an account.
Dictation, where your cursor is
Hold a key, speak, release. Clean, punctuated text lands in whatever has focus, on your PC, your Mac and your phone, with your clipboard handed back.
Free mode: talk and point
Narrate for as long as you like while you grab a region, pick a button on screen or take a selection. One note, what you said interleaved with what you pointed at, pasted where you finish and saved as Markdown.
Screen grabs, annotated
Freeze every screen at the keypress, drag a region, add arrows, boxes, text and numbered steps, and pixelate what is private. Saved as a PNG that stays editable.
Screen recording, shared as a link
Record the screen with your narration. The recording writes itself up with a title, chapters and a summary, and from your Deixis account it shares as a link anyone can open without one. The Loom you can stop paying for.
Meetings, no bot, names included
Your microphone and the other side, transcribed on your machine while the call runs. Deixis tells the voices apart and remembers the ones you name, and never uploads a voice.
Your own buttons and workflows
Add a button to the palette: it gets the note on stdin and types back whatever it prints. Send finished notes to any URL you own. Chain steps: read the text off a screenshot, translate it, summarise it, post it.
Side by side
Whisperstream and Deixis, side by side
| Row | Deixis | Whisperstream |
|---|---|---|
| Runs on | Windows 10 and 11 (any 64-bit PC from 2013 on), Mac (Apple Silicon), iPhone (as a keyboard) and Android (a dictation bubble in any app, plus meetings). One engine on all four, and audio never leaves the device. | Windows 10 or 11, 64-bit |
| Speech recognition happens | On your own device, inside the app, on every platform. Your audio never leaves it. | On your device |
| AI cleanup happens | On your device, with no provider at all. Or, if you want more, Deixis's cloud, your own Ollama, or your own key. Your choice, in one setting. | On your device or in the cloud, your choice |
| Leaves your machine | Text only, and only once you switch a cloud feature on. Never audio, never screenshots, never window titles. | Nothing |
| Account needed | None, ever, for anything that runs on your device. The cloud extras have a seven-day trial you start with an email code. No card. | None |
| Price | Free forever for everything local, meetings included. Pro is $49 a year for the cloud extras with nothing to set up. | Free tier: a seven-day trial that starts with your first successful dictation, no account needed; $49 one-time covers two devices; $29 one-time; $29 one-time for one device or $49 for two, with a 30-day money-back guarantee; when the trial ends, recording stops until a licence is activated |
| Your transcripts live | In plain files, in one folder you own and can back up. Uninstall Deixis and every note is still readable. | On your device |
| Make it yours | Add your own buttons that run a script, send finished notes to any URL you own, and switch on vocabulary packs for your field. | A dictionary of filters and replacements applied after recognition; optional AI enhancement through a downloadable local model, Ollama, or a cloud provider under your own key, with enhancement profiles you write the cleanup prompt for and route by application or website; 39 dictation languages; pin mode and spoken replies for coding-agent sessions |
Whisperstream facts from whisperstream.io, its privacy policy, features, its docs, providers, basics, languages, dictionary, storage and subprocessors, read on 5 September 2026. “Not documented” means those pages did not say.
Whisperstream is the closest thing to Deixis on this site: push-to-talk dictation for Windows, recognition on your CPU, no account, and a privacy policy that says 'No telemetry, no analytics, error reporting, crash reporting, no usage tracking'. The differences are in what surrounds the dictation. It is a paid product after a seven-day trial ($29 one-time for one device, $49 for two, with a 30-day money-back guarantee); its dictionary corrects text after recognition rather than steering it; its AI enhancement is off by default and can run on a local model it downloads (Gemma 4 E2B), on Ollama, or on a cloud provider you hold a key for, with a cleanup prompt per profile routed by app or website. A Pro licence checks in with Polar about once a week while you are online. Reviews and feature requests you choose to submit go to Supabase and Featurebase.
What leaves your machine, and what leaves theirs
Deixis publishes every byte it sends, as a table, because there is so little of it. Here is that table beside what Whisperstream's own pages say.
| What | Deixis | Whisperstream |
|---|---|---|
| Audio | Never. Recognition happens inside the app on your device; the recording, the model and the transcript stay there. | Nothing: 'Microphone audio is held in random-access memory while each push-to-talk capture is transcribed on your CPU, and is never uploaded to any server'— · — |
| The dictated text | The sentence you just dictated, your vocabulary, and your previous sentence in that fieldthe endpoint you chose · only with a cloud endpoint set, and by default only when the local rules are not enough | Nothing by default. With AI enhancement on and a cloud provider chosen, 'transcribed text longer than three words is transmitted only to the specific third-party provider you selected'the provider you chose, under your own key · only once you enable enhancement with a cloud provider |
| Context around the text | The names of the app and the field the text is going into. Never the window title, never a screenshot.the same endpoint · with that request | not documented |
| Telemetry and analytics | None. No account, no telemetry, no analytics, no crash reports. | None: 'No telemetry, no analytics, error reporting, crash reporting, no usage tracking'; aggregate counters stay on the device— · — |
| Update check | Nothing but the request; the version check happens on your PCapi.deixis-voice.com · every launch, and when you ask in Settings | A request with standard HTTP headers, which includes your IP addressWhisperstream's release server, on Cloudflare · automatically; it asks before installing |
| Model download | Nothing but the requesthuggingface.co · once, when you press Download on first run | A download request with standard HTTP headers, and no licence key or account identifiera Cloudflare endpoint · at setup for the speech model, and when you add an optional voice or local AI model |
| Account and sign-in | An email address, only if you choose to start the trialthe Deixis endpoint · when you start the trial | No account. With a Pro licence, the licence key, hostname, OS family and a product identifier go out for validationPolar, the payment provider · about once every seven days while online with a Pro licence, and when you activate or deactivate a device |
| Integrations | The finished note, and for a screenshot its file name, never the imagea URL you added yourself · on finish, only for destinations you added | not documented |
Before you switch
Before you switch
Who should not switch
If you dictate in a language other than English, stay on Whisperstream; Deixis's models are English models and that is the whole answer. Stay if you want a local language model for cleanup without running a server yourself, if you want to write your own cleanup prompts per app, or if a one-time purchase matters to you more than free-plus-subscription. On a PC, Deixis needs 64-bit Windows 10 or 11 with a processor from 2013 or later; Whisperstream asks for 64-bit Windows 10 or 11 and 8 GB of RAM. And if what you want is a dictation box and nothing else, Whisperstream is a good one; Deixis is the choice if you also want the meeting, the screenshot and the field-aware cleanup.
Making the switch
The two feel alike at the keyboard. Here is what changes.
The key
Whisperstream's default push-to-talk is Right Shift; Deixis's is the Ctrl+Win chord, chosen
because almost nothing else listens for it. Hold, speak, release. Double-tap it for free mode, which
keeps recording until you double-tap again. Settings → Hotkey captures a new chord or a single
function key if you would rather keep something closer to Right Shift.
Models
Whisperstream downloads the speech model for your language at setup, and others as you pick them. Deixis shows one Download button on first run, 142 MB, and offers a larger, more accurate model in Settings that swaps in without a restart. Both run on the CPU; Deixis needs a processor with AVX2, which is Intel from 2013 and AMD from 2015.
Dictionary
Whisperstream's filters become Deixis's filler-word list, and its replacements become Deixis's
replacements, matched at word boundaries and case-insensitively just the same. What is new is
vocabulary: terms that go into the speech model's prompt so they are heard right. Select a word,
press Ctrl+Win+D. Switch on a pack for your field in one click.
Cleanup
Whisperstream's enhancement is off by default; Deixis's cleanup runs as rules on your PC by default, with no provider. When you want a language model, Settings → Cleanup is one choice: When it helps, Every time or Never, against Deixis's cloud, a model server on your own machine, or your own key. The setting tells you what will leave before you confirm. If you wrote a cleanup prompt per app in Whisperstream, there is nothing to carry over: Deixis sends the app and field name with the request and punctuates for the destination itself. If you were using Whisperstream's local model, the equivalent in Deixis is a local model server, which you run yourself.
History
Whisperstream keeps an encrypted history with a retention window. Deixis keeps a history behind the Library tab and exports it on request; it is a file in your folder, readable by anything.
Before you ask
Questions people search for
Is Whisperstream really local?
Yes, by its own account: its privacy policy says microphone audio is held in memory while it is transcribed on your CPU and "is never uploaded to any server", and its AI enhancement is off unless you switch it on. Deixis makes the same promise the same way, and adds that no setting, on or off, can ever send audio.
Does Deixis support other languages?
No: Deixis's speech models are English models. Whisperstream lists 39 dictation languages, and if you dictate in one of the other 38 it is the right choice.
Can Deixis clean up text without the cloud?
Yes, and it does so by default: a rules engine on your PC handles punctuation and structure, and the dictionary strips filler words, with no provider at all. For a language model that stays local, point Deixis at a model server on your own machine; there is no fallback to the network if that server is down.
Whisperstream or Deixis: which is cheaper?
Deixis, for everything local: it is free with no trial period and no account, meetings included. Whisperstream is $29 one-time for one device or $49 for two after a seven-day trial. Deixis charges $49 a year only for the managed cloud extras, which are also free with your own provider key.
Which is more private?
They are close, and both are far ahead of any cloud service: neither uploads audio and neither runs telemetry. Whisperstream's Pro licence checks in with its payment provider about weekly, and its cloud enhancement, if you enable it, sends text to the provider you chose. Deixis has no licence check, and text leaves only once you set a cloud endpoint, with the app and field name alongside and never the window title.
Do I need a GPU?
No: Deixis's speech recognition runs on the CPU, on any 64-bit Windows 10 or 11 PC with a processor from 2013 or later, and on the Mac and the phone it runs on the device the same way. Whisperstream says the same of its transcription, so neither needs a graphics card to dictate.
Hold a key. Speak. Keep working.
Deixis is free, needs no account, and runs on your own PC. Everything on this page is in the current release.
v0.9.0 · 64‑bit Windows 10/11 · 46.7 MB · nothing else to install
Facts about Whisperstream checked on 2026-09-05 against whisperstream.io. Deixis facts are from its README for the current release. Something out of date? Tell us.