Desktop app
Answers on the call,
while you are still on it.
Assist listens to a live sales call and answers the questions your prospect asks, from your own documents, in a window that floats above the call.
Why it exists
The problem it was built for, as it actually showed up.
What happens on the call
Reps get asked things they are not sure of, and sometimes even for basic questions they would say “I will get back to you” to the client, especially the ones that are new to the job. That leaves a bad impression, particularly when it is something they are expected to know.
Why it is not their fault
There is a lot of information, things keep getting updated, and someone else might have discussed something with the client earlier on a call or over email. Preparing for all of that takes a lot of time and effort, and none of it helps in the three seconds you have mid-sentence.
Install it
Windows 10 or 11. The installer is on GitHub, alongside the source.
-
Download the installer
Take
Assist-Setup-0.1.0.exefrom the latest release. It is around 202 MB, because the speech and retrieval service is bundled rather than downloaded later. -
Run it, and let Windows ask
The file comes from the internet and is not signed with a paid certificate, so SmartScreen will warn you. Choose More info, then Run anyway.
-
Point it at your documents
Add the decks, one-pagers and FAQs the answers should come from. Assist indexes them locally and answers only from what you gave it.
-
Start a session when the call starts
Pick a meeting profile, and the overlay sits above your call window. Drag it anywhere that suits you.
Source and releases
The repository holds the code and every release, if you would rather read it before running it.
View on GitHub ↗Mac
Not yet supported. Currently Windows only.
How it works
Three steps between the prospect finishing a question and the answer being on screen.
Listens
A model transcribes the call as it happens. A rep voiceprint keeps the seller's own speech out of the trigger path, so your talking never sets it off.
Finds
Another model picks out the genuine questions, and retrieval pulls the matching passages from your indexed documents.
Answers
A stronger model writes the answer with its sources, in a translucent window above the call that you can drag anywhere.
Built to be wrong out loud
The guardrail is the feature, not a disclaimer at the bottom.
Every answer is graded
Each one is checked against what retrieval actually found. A weak match is labelled and told to escalate rather than dressed up as an answer.
Why that matters more here
On a live call a confident wrong answer is far more expensive than an admitted gap. The rep can recover from “the documents do not cover that”. They cannot recover from being confidently wrong in front of a client.
If something looks off
Windows is warning me about the installer
Expected. The build is not signed with a paid code-signing certificate, so SmartScreen flags it the way it flags any unsigned download. Choose More info, then Run anyway. The source is on GitHub if you would rather check it first.
It is not answering anything
Assist only answers from the documents you gave it. If the knowledge base is empty, or the question is outside it, it says so instead of inventing an answer. Add the relevant deck or FAQ and try the same question again.
It triggers on my own voice
Record the rep voiceprint in settings. It is what tells your speech apart from the prospect's, and without it the app cannot know which side of the call to listen for questions on.
Why a desktop app and not a browser tab
The two things this needs, capturing the meeting's audio and holding a window above the call, are exactly what a browser tab cannot do. The cost of that choice is an installer instead of a URL.