You talk. It listens, writes it down, and tells you what matters.
Calls, meetings, and recordings you already have.All of them transcribed, written up and filed the same way.
Three ways in for a conversation, and the same treatment for all three. Place or take the call here — a proper business phone, on any IP PBX you already have. Sit in a meeting in Zoom, Teams or Meet and it is recorded from the computer itself. Or drop in a recording made somewhere else entirely. However it arrived, it is kept, turned into text you can search, and written up — the summary, the tasks, the category — without anybody typing.
Windows, macOS and Linux. No account, no subscription, no cloud service in the middle — see how it works.
Three ways incalls, meetings, and files you already have
All localcalls, text and AI can stay on your machine
GPL v2free software, no licence per seat
30 languagesa complete interface in every one of them
How it works
From any conversation to a written summary, with nobody typing it up
Four steps. You are only involved in the first one.
The conversation arrives
Three ways in, and after this step it no longer matters which one it was: a call you place or take here, a meeting running in Zoom, Teams or Meet on the same computer, or a file you already have and drop in.
It is recorded
A call or a meeting is recorded on two channels, you on one and the far side on the other, so who said what is never in question — and the way the other party is told about it is yours to set. A file you brought in is a recording already, and goes straight to the next step.
It becomes text
The recording is transcribed on its own, turn by turn, in the language it was spoken in. Click a line and hear it. Search a year of calls by what was said in them.
It becomes a write-up
A summary, the tasks the call left you, the subject it was really about, the things worth flagging — and a box where you can simply ask the conversation a question.
Screenshots
The interface, window by window
Fifteen windows, from the summary a finished conversation leaves to the diagnostics an IT department asks for.
What it does today
The product as it stands
From the conversation arriving to the write-up it leaves behind — and the settings, the integrations and the diagnostics underneath.
Any conversation, whichever way it arrived
What matters here is what happens to a conversation afterwards, not where it came from. Three ways in, one library: the same recording, the same transcript, the same write-up, and one search across all of it.
A call placed or taken here, on any IP PBX you already have
A meeting in Zoom, Teams, Meet or anything else, recorded from this computer
A recording you already have — from a mobile, a dictaphone, another system — dropped in and treated like the rest
Nothing to sort by hand afterwards: an imported conversation is filed exactly like a dialled one
Conversations you can read, and search
Every recording becomes text, speaker by speaker, kept in step with the audio — click a line and hear the moment it came from.
Search everything you have ever recorded, by what was said in it
Whatever language the call was in
Export it as plain text or as subtitles
Your choice of transcription service — including one running on your own machine, so the audio never leaves the building
The write-up, done for you
As soon as there is a transcript, the phone reads it and files it: what the call was about, what was decided, and what it left you to do. Every one of these is a template you can open, reword and put back.
A summary of the conversation, and a one-line version of it for the list
The tasks it left you, the subjects it covered and the names that came up
Filed by itself — sales, support, personal — with your own tags, and the things you would rather hear about early
Scored, for the sales and support calls, against criteria you write yourself
Ask it a question — about one conversation, answered out of what was actually said
Translated into the language you work in
Your AI, wherever you keep it
The phone talks to the model you choose — one running on your own hardware, or a service you already pay for — and tells you what it is costing while it does.
A model on your own machine, or OpenAI, Claude, Mistral, DeepSeek, Groq and others
Every call priced, in tokens and in money
A monthly ceiling that pauses the work instead of surprising you
One button that checks that a service really answers
Meetings held somewhere else
A conversation in Zoom, Teams, Meet or anything else is recorded from the computer itself — the far side from what the machine is playing, you from your own microphone — and lands in the same library, transcribed and written up like everything else.
Starts by itself when a conversation begins, or waits for you to press the button
Any application, including ones nobody has heard of — it detects a conversation rather than matching a list of names
Still two channels, kept in step, so the write-up knows who said what
Visible the whole time it is running, and running only while there is a conversation to record
It cannot announce itself inside a meeting it does not host, so telling the room is up to you — it reminds you when capture starts
Every conversation kept
Recording runs by itself, with you on one channel and the other party on the other, so what each side said is never in doubt.
A spoken announcement, a tone, your own confirmation — or nothing at all: the rule is yours to set
Played back with both sides shown as waveforms
Old recordings are cleared out automatically, by rules you set
Delete one and everything made from it goes with it
Everything you expect from a business phone
Register as many lines as you have, run several calls at once, and move them around the way the day demands.
Several accounts and several calls at the same time
Hold, mute and the keypad, where your hand expects them
Transfer a call straight through, or speak to the person first
Conference in the people already on the line
Clear, wideband sound wherever your provider offers it
Fits the software you already run
Three ways in, from no code at all to a couple of lines of it, and each one off until you want it. An afternoon of work rather than a project — the integration page walks through all of it.
Click a number anywhere — the phone answers the telephone links your system already makes, so a number in a browser, a CRM or a document dials here without a line of code
Pop the customer up in your CRM — the phone can tell another system the moment a call rings, is answered, or ends
Drive it from your own software — your contacts, your history and the calls in progress, over a local connection that never leaves the machine
Show your IT what is wrong — a live view of the conversation between the phone and the switch, and the quality figures for every call
Yours, and private by default
Passwords live in the keychain your operating system already trusts. Recordings, transcripts and history live in a file you own. Nothing about a conversation — not a number, not a name, not a word of what was said — goes anywhere you did not send it yourself.
One small usage report a day — the version, the kind of machine, and how many calls there have been. Never a number, a contact, the address of your switch or a line of a conversation. You are shown what is in it before the first one goes, and how much it carries is yours to set
A new version announces itself when it appears — never over a call — and installs only when you say so
Dark and light, and thirty interface languages, changed without a restart
Free software under GPL v2 — read every line of it if you like
Contacts and history, beside the phone
Your address book and your call log stand next to the dialler, and a number that belongs to somebody shows their name everywhere it appears.
Every call, in and out, with what came of it
One click to call back
Kept in a file that belongs to you, not in somebody else’s cloud
Always on, out of the way
Close the window and the phone stays on the line: it sits in the tray, tells you who is calling, and answers a hotkey while you are working in something else.
Tray icon, close to tray, start with the computer
A banner for the incoming call, wherever you are
Hotkeys that work when the phone is nowhere in sight
Answers automatically when your shift — or the PBX — requires it
Sound, set up once
Choose your headset by the name your computer shows it under, ring on the speakers while you listen on the headset, and hear that it works before the first call rather than during it.
A separate device for the ringtone
Test sounds and a live microphone meter
Changed mid-call without dropping it
Only the parts you want
The product is a small core with its capabilities beside it rather than built into it. Turn one off and it is gone — no settings page for it, no menu entry, nothing of it running.
Recording, transcription, the write-up, the meeting capture, the file import, the integrations — each on its own switch
A phone that is only a phone, where that is all a desk needs
The same package either way: nothing to buy, nothing to unlock
Personal Prompt Studio
Every conversation, written up your way.
Start with 11 ready-made prompts, then change the words, the criteria and the rules to fit the way you do business. Your LLM, your key, your control.
Ready-made prompts
A summary, action items, topics, names and numbers, categories, tags, red flags, and sales and support reviews — ready as soon as transcription and AI are connected.
Your own wording
Every prompt is an instruction in ordinary language. Change what it looks for and how it answers, or duplicate it for a new purpose.
Your rules
Decide what runs after each conversation, in what order and on what condition: a sales call gets a sales review, a support call a support one.
Your software controls the calls. The softphone reports back.
The simplest way to get CTI integration up and running in minutes. A local REST API and built-in webhooks generator. Enable only when you need them. No SDK, no intermediary cloud service, and no externally exposed network listener.
You control
A localhost-only REST API lets your CRM or other applications control the softphone. Start calls and access contacts, call history, SIP accounts, and active calls.
It notifies you
Webhooks send an HTTP request to an endpoint you choose whenever a call starts, changes state, or ends. Your CRM can open the customer record before the second ring.
The interface in thirty languages, the conversations in any
English and twenty-nine others: each of them a complete interface rather than a half-translated one, chosen in the settings and changed without a restart. What language a call was spoken in is a separate question — the transcript and the write-up follow whatever was actually said.
This page opens in whichever of them your browser asks for — the selector in the header overrides it, and remembers.
What is next
What we are building next
None of this is in the product yet. Four things are under way — and a line is left open for yours.
Live transcription
The transcript appearing during the conversation instead of after it, as subtitles on the screen — with a translation for the party who does not share your language.
Live suggestions from your knowledge base
Prompts while the conversation is still going: what to ask next, what to answer, and what your own documents say about the question — driven by instructions you write yourself rather than somebody else’s.
Your own voice robot
The phone handling the conversation itself, on a script you write: answering the line, asking what it has to ask, and handing over to a person at the point where it should.
The company rather than the single desk: one knowledge base for everybody, one place to administer every seat, shared storage for the recordings and the transcripts, reports and quality scoring across a team, and the work of a department in a single view.
Download
Get AI Softphone
Free software under GPL v2. No account, no subscription, no cloud service in the middle.
What your machine needs
Taken from the packages themselves rather than from intentions, and kept in step with the product’s own requirements page.
Windows
Version
Windows 10 version 1809 (build 17763) or newer, and Windows 11
Processor
64-bit Intel or AMD
Graphics
Direct3D 11, or OpenGL 2.1
Sound
Whatever Windows itself plays through; nothing to install
Disk
250 MB, and whatever the recordings come to
Nothing has to be installed first: the C++ runtime, OpenSSL and Qt are all inside the package. On an older Windows the installer refuses to run and names the build number it found, rather than letting it fail afterwards.
Linux
Distribution
Ubuntu 22.04 LTS or newer, Debian 12 or newer, and anything of the same age — Fedora 36+, openSUSE Leap 15.5+, Mint 21+, Arch
C library
GNU C library 2.35 or newer
Processor
64-bit Intel or AMD
Graphics
OpenGL 2.1 or OpenGL ES 2.0, on X11 or Wayland
Sound
PipeWire or PulseAudio; ALSA where there is neither
Desktop
Any. The tray icon needs one with a status-notifier area
Disk
200 MB, and whatever the recordings come to
The AppImage carries its own Qt, its own codecs and its own C++ runtime, so the only thing your distribution decides is the age of its C library — and where that is too old, the image says so on the screen instead of failing in the loader. The .deb asks for one thing more: the system’s own C++ runtime from GCC 13, which Ubuntu 24.04 and Debian 13 have and 22.04 has not. On anything older, the AppImage is the one to take.
macOS
Version
macOS 14.4 or newer — the package refuses to open on anything older rather than starting and failing
Processor
Apple silicon. An Intel Mac cannot open it at all, not even through Rosetta; one package for both is still to come
Graphics
Metal, which every Mac this runs on has
Sound
Core Audio, and a permission for the microphone the system asks for once
Disk
160 MB, and whatever the recordings come to
Nothing has to be installed first: Qt, the C++ runtime and OpenSSL all travel inside the bundle. It is notarised and the ticket is stapled to the application as well as to the disk image, so there is no dialog about an unidentified developer — and no Gatekeeper setting to change by hand.
You will need a SIP account — from your provider, or from the PBX you run yourself. Recording works the moment it is installed; the transcript and the write-up need either a service you choose or a model on your own machine.
Contact
Ask a question, suggest something, or tell us what went wrong
The same form the phone carries on its About page, going to the same place. How something works, what you would like it to do, a switch you need it to talk to, a language you are missing — it all reaches the same people.
This page can count visits with Google Analytics — which pages are read, and roughly where from. That needs a cookie, so it waits for your answer.