Ledger reads your Zoom and Google Meet transcripts and extracts the decisions — not notes, not summaries — with the owner and the deadline attached. Then it tells you when a new decision quietly reverses an old one.
First 10 meetings free. No credit card. Works with the transcripts Zoom and Meet already produce.
Move the mobile app redesign ship date from October 15 to November 5.
“Then we’re moving the mobile app redesign ship date to November 5th. That replaces October 15th.”
Contradicts “Ship the mobile app redesign on October 15” — decided 14 Jul in Q3 Planning. Needs review.
Switch the notifications service datastore from Postgres to DynamoDB.
In-app chat stays cut from v1 and remains in the v2 backlog.
The decision to cut a feature lives in someone’s Notion page. The reversal three weeks later lives in a Slack thread. Nobody notices they contradict, the owner never hears the deadline moved, and six weeks on the same argument happens for the third time — because there was never a record to point at.
Notetakers made this worse, not better. A summary of everything said is not a record of what was settled.
No bot joins your call. No new meeting habits. Ledger works from the transcript your platform already generates.
Drop in the .vtt from a Zoom cloud recording, a Gemini notes doc, or paste text. Ledger detects the format and normalises it.
Every extracted decision arrives with an owner, a deadline, a confidence score, and the exact line of transcript it came from. You accept, edit, or reject.
New decisions are checked against everything already on record. Contradictions are flagged for a human — never auto-resolved.
Extracting decisions is table stakes. The value is knowing when a decision you already made stops being true — and telling the person who owns it.
“We’re going with Postgres” and “switch the notifications store to DynamoDB” share no words. Ledger compares meaning, so a reversal phrased entirely differently still surfaces.
A missed contradiction costs weeks. A false flag costs ten seconds. Ledger errs toward showing you the pair and letting you dismiss it.
A model deciding which of two decisions is “right” would be a liability. Ledger surfaces the pair with both sources and a person chooses.
A full replacement supersedes the original and links back to it. A partial change is recorded as a resolved conflict with a note, so the original still reads as the decision it was.
Each one links to the exact span of transcript it came from. When someone disputes it, you open the line.
Pulled only when someone actually accepted the work. No owner is recorded when nobody took it.
Hedged discussion scores low and routes to review. Confident extraction of ambiguity is the failure mode Ledger avoids.
One email: decisions past their deadline, decisions with nobody on the hook, and conflicts waiting for a human.
Filter by owner, project, status, or date. Find what was decided about billing in March without opening six documents.
No bot in your meeting, no browser extension, no third-party notetaker holding your recordings.
The timestamped, speaker-labelled .vtt from your Recordings tab.
Its “next steps” sections are used as hints — checked against the document, never trusted outright.
The “Take notes for me” doc, exported or pasted.
The raw timestamped transcript doc from the organiser’s Drive.
Your workspace admin must have cloud recording and transcription enabled — Ledger reads what those produce and can’t turn them on for you.
Every row is filtered by your organisation at the database level, not by application convention. A query that forgets to scope returns nothing rather than someone else’s data.
Uploads live in private object storage and are never served from a public URL. They’re fetched server-side after an ownership check.
Single-use, fifteen-minute links. Nothing to reuse, leak, or store.
Your transcripts are processed to extract decisions and are never used to train models.
Removing a meeting removes its transcript, its decisions, and the stored file.
Everyone in your organisation should be able to read the record. Charging per seat would make that expensive on purpose.
10 meetings, no card required
Start free100 meetings per month
Start free trialUnlimited meetings
Start free trialNo. Ledger never joins a call and never records anything. It works from the transcript Zoom or Google Meet already produces, which you upload or paste. Nothing new is listening.
A notetaker summarises what was said. Ledger extracts what was settled — and keeps it, so a decision made in July can be checked against one made in October. A summary has no memory across meetings; that memory is the entire point here.
Nothing reaches your record without a human accepting it. Every extraction lands in a review queue with its confidence score and the transcript line it came from. Hedged discussion scores low precisely so it surfaces for a person rather than being asserted.
They’re stored in private object storage, encrypted at rest, isolated per organisation at the database level, and never used to train models. Deleting a meeting deletes the transcript, its decisions, and the stored file.
Yes, and it will tell you so. A status update with no decisions returns an empty result rather than inventing something to justify the upload. That case is tested explicitly.
When a meeting confirms something settled earlier — “that’s decided, two weeks instead of one” — Ledger records it and marks it as previously decided, so the date it entered the record isn’t mistaken for the date it was made.
Upload one transcript and see what your last meeting actually decided. Takes about a minute.
Start free