Granola Record Audio: What the Docs Say About Capture, Storage and Deletion

ogTitle: "Granola Record Audio: Capture, Storage and the Gap in Between"

Granola records audio. It captures your microphone and your computer's system audio, sends that audio to a third-party transcription provider, and deletes the audio after transcription completes. If you searched "granola record audio" because a client asked whether your notetaker records the call, the honest answer is yes, for a period, and the recording leaves your machine.

The more useful question is what happens between capture and deletion. Granola's own documentation answers most of it, and the answer matters if you take confidential 1:1 calls as a consultant, coach, lawyer, financial adviser or therapist.

Does Granola record audio?

Yes. Granola's transcription documentation says the app "captures computer audio inputs and outputs through system audio, along with microphone audio." That's the recording step. The same page says Granola "sends microphone and system audio to a transcription provider for transcription."

Granola does not use a meeting bot that joins calls as an additional participant. The capture happens locally on your device. But "locally" describes where the capture starts, not where the audio goes. The audio is recorded, then transmitted to a transcription provider.

Granola's security FAQ says the app "does not retain audio recordings after transcription." That's a deletion promise, not a no-recording promise. The audio exists for the duration of the transcription pipeline.

Does Granola save audio after the call?

No, according to Granola's documentation. The transcription page states that Granola "does not save audio or video during calls, so users cannot access meeting audio recordings afterward." On iPhone, Granola "temporarily caches meeting audio and deletes the cached audio from Granola and third-party systems after transcription is completed."

The distinction is between a saved recording you can replay and a transient recording that exists in memory or cache while transcription runs. For a client asking "did you record our call," the second is still a recording in the ordinary sense of the word. It's just a short-lived one.

The Granola subprocessors post on this site walks through what the public docs say about where that audio goes. Granola names OpenAI and Anthropic as subprocessors, uses US-only AWS storage, and retains transcripts indefinitely by default. The audio may be deleted after transcription, but the transcript that came from it is not.

Does Granola record without asking?

Granola requires a user action to start capture. The security FAQ says users must "click to start recording" and the app "does not start completely automatically." If a meeting runs long and you don't end the session, Granola "may continue transcribing."

That's a consent question for the person running the call, not a technical one. Granola's bot-free design means other participants don't see a notetaker in the meeting. The Granola lawsuit post on this site examines Chamberlain v. Granola, a proposed class action filed on 30 July 2026 in the Northern District of California. The complaint argues that the absence of a visible bot left other participants with no way to know a recording was running. A complaint is not a ruling, and it names the exact tension: invisible capture and all-party consent laws can collide.

This is not legal advice. Consent and professional-conduct duties apply to the person running the call, regardless of which tool captures the audio.

Where does Granola's audio go?

Granola's documentation says the audio goes to a transcription provider. The subprocessor list itself sits behind a Trust Center request, but the public security FAQ names OpenAI and Anthropic. That means your client's voice, or at minimum the audio of the call, reaches third-party systems before deletion.

For a consultant with an NDA that prohibits disclosure of client information to third parties without written consent, that's a direct question. The consultant NDA post on this site explains why a cloud notetaker can be the disclosure your contract already forbids. The audio path is the disclosure, not the saved file.

Granola's architecture is bot-free, not server-free. The bot-free is not private post makes the same point about the 2026 wave of botless notetakers: removing the bot from the participant list doesn't remove the server from the pipeline.

What's the alternative for confidential calls?

The alternative is an audio path where the recording never becomes a file and never leaves the Mac. Fazit captures the call app's output and your microphone through a Core Audio process tap, transcribes both streams on-device using Parakeet via CoreML, and writes a Markdown note into your Obsidian vault. The audio exists only in a fixed-size RAM ring buffer with no write API, and it's destroyed on every exit path including errors.

That's a different claim from "we delete the audio after transcription." Deletion implies the audio existed as a durable artifact at some point. Fazit's architecture means the audio never becomes a file at all. The never-records architecture post explains why that's an invariant, not a marketing line.

The practical difference for a lawyer is privilege. The AI notetakers for lawyers post on this site covers why sending client-call audio to a vendor and its subprocessors is a third-party disclosure that invites waiver arguments. A local workflow reduces that disclosure because call content stays on the Mac.

The comparison that matters

Granola's documentation is unusually clear about what it does. That's worth acknowledging. The app records audio, sends it to third-party transcription services, deletes the audio, and keeps the transcript. For a sales call with no confidentiality obligation, that's a reasonable trade.

For a client call where the NDA, the privilege, the § 7216 obligation or the therapy consent form already constrains what you can disclose, the audio path decides the risk. The audio path comparison post on this site ranks notetakers on where a durable artifact of the conversation is created and who holds it afterward. Five of six create one on someone else's infrastructure. Fazit is the pipeline where the artifact never exists.

Sources

• Granola transcription documentation
• Granola iOS transcription documentation
• Granola security, privacy and data FAQs
• Granola setup documentation
• Fazit security and architecture
• Fazit llms.txt