What is Granola?

The meeting recorder that turns every conversation you have into searchable markdown your AI can read.

I'm in a leadership sync. The Granola window sits on the side of my screen. I'm not taking notes. I'm in the conversation, paying attention, asking the next question. When the meeting ends I have a verbatim transcript and a clean summary in my notes folder before the next call starts. By Friday the folder has every conversation I had this week. Claude can read all of it. The product doing that work is called Granola.

It is the single most important install in my AI stack and the foundational layer almost every other workflow I run sits on top of.

What it is

Granola is a Mac meeting recorder for people who hate taking notes. It sits in the background during calls (Zoom, Google Meet, in-person), captures the audio off your laptop, transcribes the conversation in near real time, and writes both a verbatim transcript and a clean summary to a folder on your machine. No meeting bot. No "this meeting is being recorded by Otter" message in the attendee list. The recording lives only on your laptop and the people in the call.

The product is at granola.ai. Available on Mac and Windows as of 2026-05. Paid product; the basic plan is $14 a month.

What makes Granola load-bearing for the operator stack is the markdown export. Other transcription products lock the output inside their own app or surface it through a clunky UI you can't grep. Granola dumps every meeting into a folder where you (and Claude) can read it as plain text. That folder is what every workflow in the Desk Theory stack runs on.

Why it matters

For a year I tried to run my week off memory. I'd glance at my calendar on Sunday and tell myself I'd remember where I'd left things with each attendee. Then Monday would land. Then I'd ask a VP "remind me where we landed on the senior hire?" in front of the rest of the team. Then I'd lose a tiny unit of trust each time. Multiply that by twenty-six meetings a week and the cost was real.

I tried three other note-takers before Granola. Fireflies. Otter. Read. None of them got out of the way the way Granola does. The bot-free recording matters; the markdown export matters; the not-having-to-think-about-it matters most of all.

Granola is the foundation everything else sits on. Without a transcript layer feeding markdown into a folder, every workflow downstream is impossible. The pre-meeting brief has no prior conversations to read. The commitment ledger has no source quotes. The pipeline radar has no signals to detect. With Granola feeding the folder, all of it becomes reachable in seconds. The cost of context-recall drops to zero.

What a good Granola setup looks like

The configuration is what turns Granola from a fancy note-taker into the meeting library your AI runs on.

• Markdown export to a dedicated folder. Configure Granola to drop transcripts into a specific folder on your laptop (e.g., ~/notes/granola/). This is the folder every workflow article in the Desk Theory stack reads from. The full setup is walked in Granola → markdown.
• Filename convention. Granola lets you customize the filename template. Use {YYYY-MM-DD} - {meeting title}.md so files sort chronologically and the title is grep-friendly when you (or Claude) need to find something later.
• Per-meeting summary turned on. Granola auto-generates a summary at the top of each transcript. Keep this on; the summary is the cheap recall layer for the days you don't want Claude to read the full transcript.
• Calendar integration. Granola pulls meeting titles and attendees from your Google or Outlook calendar. Connect it so the markdown files arrive labeled by who was in the room and what the meeting was about.
• Privacy posture. Granola records locally by default. Some plans store recordings in the cloud. Pick your posture; for sensitive meetings (board, legal, M&A), the local-only mode is worth the trade-off.

One artifact-level note: a good Granola folder feels like a journal of your business. Open it three months in and you can see every conversation you had, in order, by topic, by attendee. The first time you open it that way you'll understand what the rest of the stack is for.

Common mistakes

• Using Granola but not exporting the markdown. If the transcripts live only inside the Granola app, you've got a fancy note-taker, not a meeting library. The export to a folder on your laptop is what unlocks every downstream workflow.
• Forgetting to keep Granola running. No transcript, no leverage. Make it a default-on app the way you'd run Slack. The minute you stop running it for a week is the week you wish you had the meetings from.
• Recording without consent. Granola's bot-free recording is convenient; it is still a recording. Tell attendees you're recording when local law requires it (most US states do not, several do, most of the EU does). Same etiquette you'd run on any call.
• Mixing personal and business meetings into the same folder. If you run multiple businesses (or a business and a personal life), put each into its own folder. Headphones.com transcripts should not leak into Desk Theory queries.
• Treating the auto-summary as the deliverable. The summary is fine for a quick recall. The verbatim transcript is what the AI workflows operate on. Skim the summary; read the transcript only when you need to; let Claude read everything.

Do this next

If you don't have Granola installed yet, the install plus the markdown-export configuration is walked end-to-end in Granola → markdown. Thirty minutes. Comes back when done. Tell me in thirty days what you found in your own meeting library that surprised you. I love hearing about the first one people catch.