Answers

Does Granola send transcripts to Sonar?

No. Granola sends nothing to Sonar; Sonar reads from Granola, read-only, about once a week, through whichever of the two doors you opened. The one-click connection reads your meeting notes and has no access to transcripts at all, and a pasted API key reads the transcript lines Granola attributes to you, and only those, only if you switch spoken-voice learning on.

The direction of travel

Nothing in Granola pushes to Sonar. There is no webhook, no export and no forwarding rule. When you connect, Sonar stores a credential, and about once a week it uses that credential to ask Granola for the last seven days, reads what comes back once, keeps the moments worth a post, and discards the text. Turn the connection off and the credential goes with it; Granola never had anything of Sonar's to keep sending.

What Sonar can ask for depends on which door you opened, and the two doors are different enough to deserve separate answers.

The one-click door: notes, and no transcript access

The one-click connection goes through Granola's own MCP server, using the standard authorisation that server accepts: you approve Sonar on Granola's screen, and Sonar registers itself as a client with no shared keys anywhere. Through that door Sonar calls Granola's tools for your stored meetings, their titles, dates and notes, from the last seven days, and receives one block of text. Transcripts are never read through it.

Because that door returns notes Granola wrote, everything from it feeds your intelligence only, the moments and the questions on Home, and never your voice. Its receipt on the Settings row says how many moments came out; it cannot count notes it never enumerated, so it does not pretend to.

The key door: summaries, and your own lines if you opt in

The second door is a key you create in the Granola app and paste into Sonar, which speaks to Granola's public API. Through it Sonar reads your last seven days of notes, the five most recent, and from each the summary Granola wrote, up to 2,500 characters, plus the title and the day. That is the material for moments, and the receipt names the notes it read.

The key door can read one more thing, and only under a condition you set. Granola's API marks each transcript line with who spoke it, and lines it attributes to "me" are the note-taker's own, which is you. When "Learn voice from your sources" is switched on in Settings under Voice, Sonar reads those lines, drops every other person's line before any model sees the text, and distils what remains into notes about how you speak, with no quote longer than six words. The summaries never join that read, because a summary is Granola's words and never yours. With the switch off, which is the default, the transcript is never touched.

What is kept, from either door

The moments: a title, an angle, one line of detail and the note they came from, labelled as your Granola notes that week. The receipt of the read on the row. And the connection itself, a refresh token for the one-click door or the key you pasted, stored server-side, so next week's read runs without asking again. Never kept: the note text, the summaries, or a single transcript line. The week's material is assembled into one transient block, mined once and discarded, and your own lines, when read at all, leave with the pass. Security has the same account for every source.

What to do next

Connect Granola with one click and read the receipt.

Sign up, open Settings, Connected sources, and choose Granola. One click at Granola's screen is the lighter door; the key is for founders who want their own transcript lines to teach their spoken voice. Everything the connector reads, keeps and never keeps is on the Granola integration page.

Start free

The longer answer.

The full account is Granola integration, and what Sonar reads and keeps from every source is on Security. Also asked: Does Sonar work with just Slack and a calendar? and Can Sonar post to LinkedIn for me?