Brifo

Security & Privacy

Your meeting audio never leaves your Mac

The technical posture behind Brifo's privacy claims: capture model, data flow, threat model, and compliance roadmap.

The one-sentence summary: Brifo captures macOS system audio locally, transcribes it on-device or via our cloud backend (text only), and never stores or uploads raw audio files. No bot ever joins your meeting.

Capture model

Brifo uses macOS Core Audio Taps (introduced in macOS 14.4) to capture system audio. The sound your Mac is already playing during a meeting. This works for Zoom, Google Meet, Microsoft Teams, Slack Huddles, Discord, FaceTime, and any other app that plays audio through macOS.

Critically, no bot or external attendee joins your meeting. Other participants see only the people they expect to see. There is no calendar integration that adds a recording bot, no browser extension that intercepts traffic, no Zoom app that needs admin approval.

Data flow

  • Audio: captured locally on your Mac. Never uploaded. Discarded after transcription.
  • Transcript (text): sent over TLS 1.2+ to Brifo's API on Vercel for AI processing (note generation, action item extraction).
  • Generated notes, action items, follow-up drafts: stored in MongoDB Atlas (encrypted at rest and in transit). Only your authenticated account can access them.
  • Calendar metadata: if you connect Google Calendar, we read event titles, times, and attendees (read-only scope) to match captured meetings to events.

What we don't do

  • We never upload raw audio.
  • We never read your email, Drive, or other Google data beyond calendar metadata.
  • We never train AI models on your data.
  • We never sell data to advertisers or third parties.
  • We never expose meeting content to other users.

Third-party processors

We use OpenAI's API for AI note generation. Per OpenAI's API data usage policy, OpenAI does not train on data sent through the API. Transcript text is the only payload sent.

We use MongoDB Atlas (encrypted) for note storage. We use AssemblyAI for transcription (text-only payload sent to AssemblyAI from our backend; raw audio never reaches them).

Threat model. What we protect against.

  • Cloud audio breach. Impossible because we never store cloud audio.
  • Account compromise. JWT 30-day expiry, OAuth-only sign-in (Google).
  • Transit interception. TLS 1.2+ end to end.
  • Malicious AI training on your data. Covered by OpenAI's no-train API policy plus our own commitment.
  • Insider access. Production database access is restricted. No employee browses user data without an explicit support request from you.

Compliance status

  • SOC 2 Type 2: in progress. Expected end of 2026.
  • HIPAA: not yet certified. The architectural posture (no cloud audio) supports many HIPAA workflows but consult with your compliance team.
  • GDPR: compliant by design. You can delete any meeting at any time, account deletion within 30 days on request.

Security disclosure

If you find a security issue, please email hello@brifo.in with a clear description and reproduction steps. We respond within 48 hours and credit your disclosure publicly (with your permission) once resolved.