You caught most of it. It caught all of it.
Their line, the translation underneath, and a reply in the language you need. The translation lands before the suggestion, so you are never waiting.
Free for a month · no card · Windows · nothing joins your call
“Necesitamos pago a 60 días y un tope de responsabilidad igual al valor del contrato.”
“We need 60-day payment terms and a liability cap equal to the contract value.”
“Trabajamos a 30 días; 60 es posible a partir del segundo contrato. El tope por el valor del contrato nos sirve — ya está en nuestra plantilla.”
The nuance you miss is the one that ends up in the contract.
Asking a third time
Feels worse than nodding, so you nod — and agree to something you had not planned.
Terms go by in passing
Mentioned once, in a sentence you half-followed, and binding later.
The pause reads as doubt
Hunting for a word looks like hesitation about your position, not about vocabulary.
Hears, translates, answers.
It hears the call
Speech is recognised on your own machine, over Zoom, Meet, Teams or a phone on speaker. Nothing joins the meeting.
The translation comes first
A separate, faster request than the suggestion — so you understand the sentence before the answer is ready.
The reply is in their language
And built from your documents: your payment terms, your discount limits, your contract template.
Interface language, answer language and recognition language are three separate settings. Fixing the recognition language when you know it is more accurate and about a third of a second faster than automatic detection.
A translator does not know your terms.
“We work with 30-day payment terms.” — an accurate translation of a sentence you had to compose yourself, under time pressure, in a language you are not fluent in.
“Net-30 is standard; 60 is possible from the second contract. The liability cap at contract value is already in our template.”
Audio never leaves the machine.
Not recorded
Audio lives in memory for about a minute and is overwritten. Nothing is written to disk.
Not uploaded
Recognition runs locally, so negotiations are not sitting in someone's transcription cloud.
Text is sent
The words of the remark go to a language model — that is what produces the translation and the answer.
$39 per month, one plan.
Free for the first month while it is in beta, and no card. Payment afterwards is handled by Gumroad, which acts as merchant of record.
Negotiating as a team? The team plan is $69 per seat and shares one knowledge base.
A key arrives in your inbox.
Free for 30 days. Windows. The download link and setup steps come with the key.
Before you ask.
Which languages?
Recognition is tuned and measured for English and Russian; other languages work through automatic detection with lower accuracy. Translation and answers cover the languages the model handles, which is most of the ones you would negotiate in.
How is this different from a translation app?
It listens to the call itself, and it answers from your documents rather than translating a sentence you had to invent first.
Will the other side notice?
Nothing joins the call and no bot appears in the participant list. The window is on your screen only.
How fast is it?
The translation typically appears about a second after the remark ends, the suggestion a second or two later. It is deliberately ordered that way — understanding first, answer second.
Does it work without internet?
No. Recognition is local, but translation and answers come from a language model over the network.