Features
Everything below follows from that. If a thing cannot be traced back to something said on the call, Overtone will not present it as though it can.
Everything that has to happen while you are listening, without taking your attention.
A camera button sits among the venue’s own controls, so it works in fullscreen. Each capture is stamped with the exact playback position and stored against the call.
One keystroke stamps the current moment. Notes stay in call order and click back to the recording.
Overtone reads the transcript the venue publishes, separates prepared remarks from Q&A, and attributes speakers.
Indexing happens when you ask for it. Opening a page does not spend your allowance or index anything.
The part that has to be defensible.
Every claim carries a marker that resolves to a timestamp, a speaker and the quote it came from.
Markers the model produces that do not resolve to a retrieved source are removed before anything is stored.
A numeric claim with no source is shown to you as unsupported rather than quietly kept or silently deleted.
Semantic search catches the paraphrase; full-text catches the rare exact terms that decide a call — “350 basis points”, “Section 232”.
Ask a question across every call you have captured for a company, not just the one on screen.
What management’s language says that the numbers do not.
Occurrences per 1,000 words of management speech, so a long call is not automatically a hedged one. Shown with the phrases behind it.
Flagged from three independent signals: explicit deflection, low topical overlap between question and answer, and a requested quantity that never arrives.
The phrases they started and stopped using, and how confidence and hedging moved against the previous call.
Confident language minus negative language, measured the same way each quarter so the trend is comparable.
The part that compounds. One call is a note; twenty quarters is an edge.
The names you follow. Calls group by ticker, which is what makes quarter-over-quarter possible.
Write what you believe, then tag notes to it as you listen. Evidence accumulates across quarters in one place.
Ticker, fiscal quarter and year are read from the title on capture, and you can correct any of it.
Transcripts are per customer and never pooled across workspaces.