Voice input in your product, without building it
Deixis partners put voice input into their own web apps without building any of it. You add one script and one element, and the text field you choose gets a mic. A user whose Deixis is running and connected to your site dictates on their own computer, at no cost to you, and their audio never leaves it. Everyone else, phones included, dictates through Deixis's cloud, which you prepay by the minute and which says so beside the mic on every take.
This page is the whole programme: what your users see, what it costs, what you earn, and how to join. Go straight to the docs; the live demo opens with the next release.
Voice input in your product, without building it
Three lines
<script src="https://deixis-voice.com/sdk.js" data-key="pk_live_…" async></script>
<textarea id="reply"></textarea>
<deixis-mic for="reply"></deixis-mic>
That is a working mic. The words land at the caret of the field it names, a textarea, an input
or a contenteditable, through the field's own value setter, so a React form sees them as typing.
There is no server code on your side, no speech-provider account, and no secret to keep: the
publishable key sits in your page and works only on the sites we set up for it. Which way each press
goes, and what to say when something goes wrong, is the element's job, not yours. The
docs have every attribute, event and styling hook.
Two routes, decided on every press
On the user's computer, with Deixis. When Deixis is installed, running and connected to your site, the mic records through it. Recognition runs on the user's own processor, the audio never leaves their computer, and nothing reaches us. Your page gets the words. These takes cost you nothing, and the line beside the mic says "On this computer · Deixis".
Through Deixis's cloud, for everyone else. A user without Deixis, or whose Deixis is not running or not yet connected to your site, records in the browser. The recording goes through Deixis to Whisper, an open speech model, on Cloudflare Workers AI, and the text comes back. Nothing is kept: not the recording, not the words. A Cloudflare check runs while the person speaks, so your minutes are spent on people, not scripts. The line "Transcribed in the cloud by Deixis" sits beside the mic on every one of these takes, with a short explanation behind it and the offer of Deixis.
Phones, tablets, Linux and ChromeOS always take the cloud route and are never offered an install, because Deixis takes no dictation for web pages there. Without a key, those devices see no mic at all, and a computer is offered Deixis instead.
What your users see
Your field, a mic button that wears your page's font and colours, and a short line beside it while it works. That line is the part we insist on. On a cloud take it always says so, and your stylesheet cannot hide it: you can restyle it, and keeping it visible is a condition of your key and of any commission.
The offer is the other part. Behind the cloud line, and in place of the cloud route once your minutes run out, a user on Windows or a Mac is offered Deixis, free, with a download link that carries your site. On Windows, the copy they install arrives knowing where it came from: on first run it shows one card with your site's name on it, Allow or Not now. For a partner on Deixis's own list, that card wears a Deixis partner badge and starts with the microphone ticked, so one Allow is the whole setup, and the user can still untick it. The badge says we know who runs the site, not that we audited it, and partners join the list with a Deixis release. On a Mac the installer cannot carry your site, so the user adds it in Deixis's settings, and the mic walks them through it.
What it costs
- 10,000 free minutes, once, when we set you up. They last a year.
- Then $0.003 a minute, prepaid in packs of $20, $50 and $100: about 6,666, 16,666 and 33,333 minutes. Each pack lasts a year from the day you buy it.
- Charged by the second, and only for takes that came back with words. A take nobody spoke into, or one the model could not read, costs nothing.
- A daily ceiling on cloud minutes, set with you, so a spike cannot run on unchecked.
- Topped up by email. We write when a fifth of your last pack (or of the free minutes) is left, when it runs out, and 30 days before credit expires, each time with a one-click link to buy a pack.
- Your users never see a balance. When yours reaches zero, the cloud route pauses: on Windows and Mac computers the mic offers Deixis, and phones, tablets, Linux and ChromeOS show no mic, for up to an hour after you top up.
Every one of those minutes is a minute somebody spent without Deixis. Which brings us to the part of the programme that pays you back.
What you save, and what you earn
A user on Deixis stops costing you cloud minutes, whether or not they ever pay us. From the day they connect it to your site, their takes on your pages run on their own computer whenever Deixis is running. That saving comes first, because it is yours from the first user who moves.
Commission comes second. When a user who came to Deixis through you subscribes, you earn 20%, for life, of every Pro and Team payment they make, provided their first payment comes within three months of their being linked to you. It is a balance of its own, apart from your cloud credit: turn any of it into cloud credit whenever you like, or take it as cash once $50 or more is available, paid quarterly. The partner terms say exactly who counts as yours, how refunds work and when it is paid.
Joining
Partners are set up by us; there is no sign-up form. Write to support@deixis-voice.com with what
you are building and the addresses of the sites the mic will be on. We set up your publishable key,
the sites it works on, the Cloudflare check it runs, up to 36 words the cloud should expect (product
names, people, jargon) and your daily ceiling, and we add your free minutes. Until then, the
data-simulate attribute lets you build and test against each state of the mic with no key at all,
one scenario per value.
Before you ask
Questions partners ask
Does Deixis become a data processor for us?
On the cloud route, yes: the recording passes through our service to Cloudflare Workers AI and the text comes back, so list Deixis among your service providers. Neither the recording nor the text is stored or logged, and Cloudflare's terms rule out training on them. On the local route nothing reaches us at all. If your rules keep us out of your chain of processors, the cloud route is not for you today; the local route still is.
What does our Content Security Policy need?
script-src https://deixis-voice.com https://challenges.cloudflare.com, frame-src https://challenges.cloudflare.com for the Cloudflare check, and connect-src http://127.0.0.1:47831 http://127.0.0.1:47832 http://127.0.0.1:47833 https://api.deixis-voice.com. Current browsers need
nothing added to style-src.
Does it work with React?
Yes. <deixis-mic> is a custom element, so it renders like any tag, and the words go in through the
field's own value setter followed by an input event, which is what a controlled input listens to.
For an editor that keeps its own model of the text, cancel the deixis-transcript event and insert
the words yourself.
What about phones, tablets and Linux?
With your key they take the cloud route every time and never see an install offer: Deixis takes no dictation for web pages on a phone, and there is no Deixis for Linux or ChromeOS. An iPad counts as a tablet, even though it reports itself as a Mac.
Can we test it before we have a key?
Yes. data-simulate on the script tag plays one scenario with a fixed sentence, with no microphone
and no request: the ordinary path, the offer, a cloud take, credit used up and the rest, one per
value, so you change the value to see the next. It works in a browser and in jsdom, so it fits your
own test suite. The live demo, on a key of ours, opens with the next release.
Keep reading
Related pages
The live demo of the drop-in mic opens with the next release
The live demo of the Deixis drop-in mic opens with the next release; until then the partner programme and the docs say how it works.
Voice input for your product, on your user's own computer where it can be
Voice input for your web app in three lines. With Deixis running and connected to your site, audio stays on the computer; otherwise, a cloud route says so.
Here is every byte that leaves your machine, for each tool
Every byte Deixis sends off your machine in one table, and the same table filled in for other dictation tools from their own documentation, dated.
See it on your own page
The docs have the three lines, and data-simulate plays the mic's states with no key, one scenario per value. The live demo opens with the next release.
These are Deixis's own terms and prices, as they stand on the date at the top of this page. A question about them? Ask us.