Create Structured In-Person Meeting Notes Without a Bot

• G

  Granola

  AI Productivity & Meetings

  Meeting notes app that captures device audio locally, with no bot joining the call

  Free tier◦Documented
• J

  Jamie

  AI Productivity & Meetings

  Bot-free meeting notes app with EU hosting and audio deleted after transcription

  Free tier◦Documented

1. 1

   Step 1: Choose a permitted capture method

   Select a bot-free desktop or local capture method that fits the meeting environment and organizational policy. Check microphone access, storage location, retention, participant visibility and whether the tool sends audio or text to external services. Run a short test before the meeting without recording confidential content. The step succeeds when the organizer knows where the audio and transcript will be processed, who can access them and how to stop or delete capture if consent is withdrawn.

   Tool: Granola

2. 2

   Step 2: Record an explicit consent statement

   Tell participants what will be recorded, why, where the result will be used and how long it will be retained. Record or document their answer according to local policy. If a participant declines, stop capture and use manual notes. Do not assume that the absence of a visible meeting bot equals consent. Verify this step by ensuring the consent record precedes substantive discussion and that the participant list reflects anyone who joined after the meeting started.

   Tool: Jamie

3. 3

   Step 3: Generate a traceable recap

   Use the complete transcript and the traceable-summary prompt. Ask for purpose, decisions, supporting passages, unresolved questions and important qualifications. Keep unclear names or inaudible sections marked rather than repaired from context. A recap should compress the discussion without changing what was agreed. Verification requires checking every decision, number, date and named responsibility against the transcript. If the transcript has no stable timestamps, cite short source excerpts instead.

4. 4

   Step 4: Extract and confirm action items

   Run the action-item prompt separately from summarization so tasks do not disappear inside prose. Record the exact action, owner, deadline, dependency and source. Leave missing owners and dates unresolved. Ask the relevant participant to confirm ambiguous assignments, especially when the transcript confuses speakers. The step passes when every task is independently actionable or clearly flagged for clarification, and when proposals or suggestions have not been promoted into commitments.

5. 5

   Step 5: Publish only the reviewed version

   Run the verification prompt on the recap, compare the output manually and keep a correction log. Remove sensitive transcript excerpts from the shareable version when they are not necessary for accountability. Share the reviewed notes with participants and provide a correction path. Keep the raw recording only for the approved retention period. Success means the distributed record identifies its source and review status and participants can challenge a mistake without editing the source transcript.

• Visit Granola ↗
• Visit Jamie ↗

• Create a traceable document summary · tested on OpenAI GPT-5 (Codex) — editorial dry run
• Extract accountable action items · tested on OpenAI GPT-5 (Codex) — editorial dry run
• Verify a document summary · tested on OpenAI GPT-5 (Codex) — editorial dry run

• OutputIllustrative scenario: consent, recap and correction log

  The synthetic conversation contained 2 decisions, 3 actions, 1 uncertain speaker label and 1 later correction. The final record preserved the uncertainty, applied the correction in the reviewed version and kept the raw source unchanged.

• DatasetTesting limitation

  Synthetic transcript and consent statement only. Executed in Codex on 2026-10-03. No live microphone, Granola workspace or Jamie account was used.

✓Last verified Oct 3, 2026