Runs on your machine · no account
Hold a key.
Speak. Keep working.
Deixis is voice input for people who live in a keyboard. Hold Ctrl Win, say the sentence, release — and clean, punctuated text lands at the caret of whatever has focus. Your editor. Your terminal. A ticket field in a browser tab.
Speech recognition happens in‑process, on your CPU. There is no upload, no account, and no window to switch to.
v0.9.0 · 64‑bit Windows 10/11 ·
46.7 MB · nothing else to install
free forever, with no account — see pricing
The one gesture
Three seconds, start to finish.
Everything else in Deixis is a variation on this. There is no record button, no review step and no paste.
Hold the chord
The panel appears beside your cursor and arms. Nothing is recorded until it does, and the keys stay yours — release and the letters are letters again.
CtrlWin
Say the sentence
The level moves with your voice, so you can see it is hearing you without looking away from what you are writing. Fillers, stutters and false starts come out.
−12 dB · 00:03.4
Release
The text is typed at the caret and your clipboard is handed back untouched. Median round trip on a laptop, end to end:
1.42 s
Double‑tap instead of holding and it keeps listening until you tap again — for the paragraph you have not finished thinking yet.
free mode →Deixis in four numbers
1.42s
Median round trip
Speech to text at the caret, measured end to end.
0
Bytes of audio uploaded
Recognition is in‑process. Audio never leaves the machine, ever.
1
Network call of its own
The update check, and it sends nothing but the request. The speech model downloads only when you press Download.
—
Accounts required
None. Deixis never asks you to sign in to dictate.
Hold the chord, press a letter
Seven modes, one muscle memory.
While the chord is held, the mode keypad appears and its letters go live for five seconds. Press the letter or click the chip — then the letters are letters again.
Dictation
Cleaned text at the caret, in whatever app has focus.
Free mode
Narrate for as long as you like, collecting things you point at.
Screen grab
Freeze the screen, drag a region, quick‑save or annotate it.
Meeting
Records your mic and the system output, transcribed on this machine.
Video
Record the screen with narration, then share it as a link.
Translate
The selection translated in place, with copy and replace.
Vocabulary
Send a selected word straight into the dictionary so it is never wrong twice.
Your own
Add a script to the keypad and the toolbar. It gets the note on stdin.
Free mode
Talk and point at the same time.
The product is named after the words whose meaning depends on where you are pointing — this, that, here. Free mode is that idea made real: narrate for as long as you want, and while you talk, grab a region, pick a UI element, or take a selection. The finished note interleaves what you said with what you pointed at.
Meetings
An hour of talking, transcribed on your own CPU.
Deixis records your microphone and the system output, so both halves of a call are captured, and transcribes the whole thing locally. The recording, the transcript and the audio are files in a folder you own. Notes on top of it are one optional call away.
Sharing
Send a link instead of a file.
A grab or a screen recording becomes a page with a title and a summary written for you — the thing a colleague can open and understand without you narrating it a second time.
Privacy, in full
Here is every byte that leaves your machine.
Not a policy page — the actual list. Deixis has no account, no telemetry and no analytics, so this table is short enough to print. Audio is never on it.
| What | What is sent | Where | When |
|---|---|---|---|
| Update check | nothing but the request — the version comparison happens on your machine | api.deixis‑voice.com | every launch |
| Speech model download | nothing but the request | huggingface.co | first run, when you press Download |
| Cleanup | the transcript, your vocabulary terms, and where it is going: the focused app, the control and field name, and your previous utterance in that field | the endpoint you configure | only when rules are not enough |
| Translate | the selected text and your vocabulary terms | the same endpoint | when you press the key |
| Meeting notes | the meeting transcript | the same endpoint | at the end of a meeting |
| Integrations you add | the finished note, and for a grab its path and filename — never the image | the URL you enter | on finish |
Never sent, under any setting
Audio. Screen grabs. Recorded meeting WAVs. Window titles. There is no setting that turns those on, because there is no code that reads them for the network.
The bottom four are off until you configure them
With no endpoint set, dictation falls back to the local rules engine and Deixis works entirely offline. The update check is the one call a fresh install makes on its own.
One setting decides all of it
Where does the model run?
A language model is a second pass on top of the local rules, and it needs somewhere to live. Deixis asks that once, in plain words, and shows you what each answer means before you pick it. Two of the four answers send nothing at all.
The table above is the whole of it for the desktop. The full privacy policy covers the phone apps too — including exactly what the Android dictation bubble reads, and what it never does.
No database, no sync, no lock‑in
Everything is a file, in one folder you can back up.
Your notes are markdown. Your settings are TOML you can edit in a text editor and reload without restarting. Your history is JSONL. If you uninstall Deixis tomorrow, everything it made is still readable.
%USERPROFILE%\.deixis\
- config.toml
- every setting, integrations and buttons
- dictionary.toml
- vocabulary, fixes, filler words
- notes\
- free‑mode notes as markdown, with their references
- captures\
- grabs as PNG, still re‑editable
- sessions\
- meetings: audio, transcripts, notes
- history.jsonl
- every dictation, with what it cost
Deixis on your phone
A second microphone for the machine you are sitting at.
Hold, speak, and the words land at the caret on your PC — from the other side of the room, or standing at a whiteboard. Speech is still decoded on your own machine; the phone is a microphone with a screen, not a new place for your voice to go.
Make it yours
It ends at a shell command, not a plugin API.
Deixis does not have an extension marketplace. It has stdin, stdout and your own URLs, which is a great deal more useful.
Custom buttons
Add a script to the keypad and the toolbar. It gets the finished note on stdin, and whatever it prints is typed back at the caret.
[[buttons]] label = "File a Jira" key = "J" run = "pwsh jira.ps1"
Webhooks
Every finished note, translation or grab can POST to a URL you own. Obsidian, Linear, a Slack channel, your own inbox — nothing goes through us.
Vocabulary packs
Five packs ship with it, and a pack is one TOML file — so a team can keep its product names, service names and people in one shared library.
- dev‑stack
- cloud
- medical
- legal
- finance
Pricing
A subscription buys convenience, not capability.
Every cloud feature is open to anyone with their own provider key, free, forever. Pro exists because obtaining a key is a ten‑minute detour most people never come back from.
Free
$0
forever, and with no account at all
Your own key
$0
plus whatever your provider charges you
A supported path, not a loophole.
Pro
$49/ year
roughly half what paying monthly costs
No card. An email address and a code, when you choose to start it.
In US dollars, before local sales tax or VAT — added at checkout where it applies. Card or PayPal, sold through Polar as our merchant of record, so the receipt comes from them. Cancel any time and keep what you paid for to the end of the period.
Seven days, and they start when you say so.
The trial does not begin the day you install. You start it when you are ready — an email address, a six‑digit code, and seven days of everything on the Pro list. No card, and nothing to cancel.
When it ends, only the cloud extras stop and everything local keeps working exactly as before. Nothing is taken away, and nothing stops working offline.
| What it does | On the trial? |
|---|---|
| Cloud cleanup of a dictated utterance | seven days |
| Meeting notes | seven days |
| Translating a selection | seven days |
| Share links, and the meeting archive | seven days |
| Dictation, transcription, recording, your dictionary | yours, always |
| Anything that runs on your machine | yours, always |
Before you ask
The honest answers.
Will it run on my machine?
64‑bit Windows 10 or 11, on a CPU with AVX2 — Intel Haswell or AMD Excavator, 2013 and later. That is a hard floor, not a recommendation: speech recognition is compiled against those instructions, so an older CPU stops with an error rather than running slowly. Nothing else needs installing.
macOS or Linux?
Not yet, and there is no date to give you. The parts that make Deixis feel like part of the OS — the global chord, the caret, reading the focused control — are the least portable code in it, so a port is a rewrite of the half that matters rather than a build target. If it happens it will be announced with a release; the release notes list in the footer is the place it would show up.
Does it work offline?
Completely, and that is the default. Clear the endpoint and dictation falls back to the local rules engine; meetings still record and transcribe. Only cleanup, translate and meeting notes ever want a network.
How is this different from Windows dictation?
It punctuates and structures what you said instead of transcribing it literally, it knows the app you are dictating into, it learns your vocabulary, and it never takes focus away from your work. And it does not need a network.
What happens to my data if I stop paying?
Nothing. Your notes, history, grabs and meetings are files on your disk and stay readable. The cloud features stop; the app does not.
Is it open source?
No — Deixis is proprietary. The proxy contract is documented, though, so you can point it at your own deployment and audit exactly what leaves.
Stop typing what you could say.
Free forever, no account, and the audio never leaves your machine. Install it and hold the chord — you will know inside a minute.
v0.9.0 · 64‑bit Windows 10/11 · 46.7 MB