The point of a meeting is what happens after it. This tool turns a raw transcript into the outputs that actually matter — decisions made, action items with owners, and key discussion points — using your own OpenAI or Anthropic key, entirely in your browser.
Why structure beats summarisation
A general meeting summary tells you what was discussed. That is not what drives follow-through. What drives follow-through is a clearly attributed action item with an owner and ideally a date — and the ability to check next week whether it was done.
The extraction prompt this tool builds is designed for accountability, not comprehension. It tells the model to:
- Extract decisions as discrete statements, not summaries of discussion.
- Extract action items as verb-first tasks with named owners wherever a speaker accepted responsibility.
- Capture due dates only when explicitly stated — not inferred, not estimated, only when a participant said a specific date or timeframe.
- Leave owner blank and mark as “Unassigned” where responsibility was unclear, so the gap is visible rather than hidden.
How it works
Choose a provider and model, paste your API key, drop in the transcript, and pick an output format. The tool builds a structured extraction prompt that asks the model to attribute each action item to a named owner where possible, capture due dates only when explicitly stated, and avoid inventing decisions or tasks. It sends one direct request to the provider and returns the result in Markdown for you to copy.
For Anthropic, the request includes the official direct-browser-access header so it works straight from the page.
Output formats
- Structured — labelled sections: Decisions, Action items (owner + task + due date), Key discussion points. Best for pasting into a project tracker or wiki.
- Recap email — a short, friendly summary message ready to send to attendees. Best for sending within an hour of the meeting while context is fresh.
- Formal minutes — attendees, agenda topics, decisions, actions, and next steps. Best for board meetings, client meetings, or any meeting with a formal record requirement.
Getting the best results
Include speaker labels. Transcripts with “Alice: we need to get the vendor quote by Friday” allow the model to attribute actions to Alice. Without labels, actions get marked Unassigned.
Clean up crosstalk. Most transcription tools mark crosstalk or incomplete sentences as [inaudible] or [crosstalk]. Leave those markers in — the model skips them correctly. Removing them manually before processing is not necessary.
Handle long transcripts in chunks. For meetings over 90 minutes, split the transcript into sections and summarise each one separately, then combine the action-item lists. Sending an extremely long transcript in one request can exhaust the model’s context window and produce a truncated output.
Tips
- Paste the transcript with speaker names intact so owners are attributed correctly.
- For long all-hands recordings, summarize in chunks and combine the action-item lists.
- Always confirm owners and due dates against your own recollection — AI extraction is a strong first draft, not a system of record.