What's inside Pricing FAQ Blog Documentation Sign in

A knowledge base that shows up in the conversation

Most internal knowledge fails the same way: it is written down correctly and then never read at the moment it would have mattered. Here the notes are mixed into the answer while the call is running — which is the only moment they were ever going to be useful.

Included on the paid plans · 50 MB on Pro, 100 MB on Max

Settings
Sources
MicrophoneMacBook Pro
MonitorEntire screen
Display
Opacity100%
Font size15px
Answers & context
Auto-answers
Answer my own voicesolo practice, without the other party
Knowledge basemixes in your notes
System Designstructured answer with diagrams
Teleprompter
Teleprompter modeThe overlay turns into a prompter — same panel, no extra window
Live Translation
Live Translation
Capture & privacy
Stealth modehidden during screen sharing
Multi-screenshotbatches screens into a single send
Noise suppressionapplies instantly
App Language
App Languageen
Quit app
0/0

Let's start with Q3 — how did conversion move?

Conversion is up 14% — let me pull up the funnel.

And the cost per lead?

Cost per lead is €12.40 — 9% under plan. Most of the gain came from organic.

Not a video — this is Whisperer itself, and the panel really answers. The microphone starts only when you press it.

Mechanism

Retrieval, not a bigger prompt

Nothing is dumped wholesale into the model. For each question, the fragments that actually relate to it are selected and mixed into the answer alongside the running transcript and your profile. That is what keeps a large knowledge base from making answers worse instead of better.

Notes link to each other with wiki-links, and those links form a graph — so a note brings in the ones it depends on rather than sitting alone.

What to put in

Five notes beat fifty documents

The common failure is uploading everything and concluding that retrieval does not work. What works is a small number of notes that answer questions you are actually asked out loud, written in the words people use when they ask them.

  1. The answers you repeat

    Pricing floors, security posture, what is included and what is billed separately. If you have explained it three times this month, it belongs here.

  2. Your own record

    Projects with outcomes and numbers, two or three stories per competency. This is what makes an answer sound like you rather than like a model.

  3. What the meeting map produced

    Conclusions worth keeping can be sent from a meeting into the knowledge base, so the next conversation starts where the last one ended.

Limits

Two things it is not

  1. First Not on the free plan

    The knowledge base opens on Start, Pro and Max. Free gives 60 minutes of recognition a month, which shows the meeting side but not the part that makes answers specific to you.

  2. Second Not a document archive

    Storage is 50 MB on Pro and 100 MB on Max, which is a great deal of text and not many scanned PDFs. It is built for notes that answer questions, not for keeping every file the company has ever produced.

Questions

What people ask first

What is RAG, in one sentence?

Instead of hoping the model already knows your facts, the relevant fragments of your own notes are retrieved for each question and given to it along with the question.

Does a bigger knowledge base give better answers?

Only up to a point, and past it the opposite. Retrieval selects what looks relevant, and a base full of near-duplicates gives it more ways to select the wrong thing.

Who can see my notes?

You. External AI clients reach them only through MCP, only if you approved that area, and only alongside the meetings you explicitly shared.

Can I put meeting conclusions into it automatically?

Nodes from a meeting map can be sent into the knowledge base as a note. It is a deliberate step rather than an automatic one, for the same reason tasks are not created automatically.

What happens if the answer is not in my notes?

The model answers from general knowledge, and the answer is correspondingly general. That gap is the most reliable signal of which note to write next.

Write five notes before the next call

The questions you already know you will be asked. Then take the call and watch whether the answer that appears is yours or anyone's — that difference is the whole feature.