Privilege Note

AI meeting notes that never leave your firm.

Transcripts, speaker labels and summaries, produced entirely on your firm’s own server. No copy ever reaches us or any AI company.

Request a demo

AI notetakers are already everywhere in business

  • 1 in 3

    U.S. workers say an AI notetaker has sat in on their meetings.

  • 25M+

    people use Otter alone, across more than a billion meetings.

  • 3×

    year-over-year growth in monthly users of Zoom’s AI Companion.

Much of the business world has adopted it. Law firms, where meetings are often privileged, largely can’t.

Sources: Kolmogorov Law / Pollfish survey of 500 U.S. workers, July 2026 · Otter.ai, March 2025 · Zoom Q4 FY2026 results

Lawyers want AI meeting notes. Most firms can’t say yes.

  1. 01

    The tools send the conversation out

    Popular notetakers like Otter, Granola and Fireflies process meeting audio on a vendor’s or AI provider’s servers.

  2. 02

    Client conversations leave the firm

    Once a privileged discussion sits on someone else’s infrastructure, the firm no longer controls who can reach it.

  3. 03

    So firms keep the category switched off

    Cautious firms bar AI notetakers from client matters, and lawyers go back to taking notes by hand.

But courts are starting to answer the question

U.S. v. Heppner · S.D.N.Y. · Feb. 2026

A defendant’s exchanges with a consumer AI platform were protected by neither privilege nor work product. The court pointed to the provider’s policy allowing retention and disclosure.

Ambriz v. Google · N.D. Cal. · Feb. 2025

Letting wiretap claims over an AI call-transcription tool proceed, the court looked at whether the vendor was capable of using the data, not whether it did.

In re Otter.ai Privacy Litigation

Wiretap and biometric (voiceprint) claims survived a motion to dismiss in Aug. 2026.

Chamberlain v. Granola

Proposed class action filed July 2026 on a similar theory.

Plain-English summaries, not legal advice. Ambriz was decided at the pleading stage. Status as of October 2026.

A promise is not an architecture

Data held by: vendor

Cloud notetaker

Audio goes to the vendor’s cloud. A privacy policy and a contract are what protect it.

Data held by: vendor

“Private instance”

A dedicated copy, still in the vendor’s cloud and under the vendor’s control.

Data held by: your firm

Privilege Note

Runs on your firm’s server. We never receive audio, transcripts or notes.

“Zero data retention” is a promise. Never having custody is an architecture.

How it works

Inside your firm · your office or your own cloud account

Attorney’s laptop

Records the meeting and sends it to your server.

Your firm’s server

AI runs here
  1. Transcribes the audio
  2. Labels who said what
  3. Drafts the summary and action items
  4. Stores everything, encrypted

Your attorneys

Review and edit, filed under the matter.

One Linux server · No GPU required · Air-gap ready

0 outbound connections

Outside your firm: no connection

AI companies

Never contacted.

0 external AI calls

Privilege Note (us)

No copy ever reaches us.

License checked offline

Usage tracking

Nothing reports back.

Telemetry: none

Every meeting, summarized and filed by matter

The meeting page of the Privilege Note app: a strategy call filed under a client matter, with four named speakers, an audio player, and a summary followed by key points, decisions, questions and next steps.
Illustrative example. The matter and people are fictional.

  • Named speakers

    Names display for transcript diarization. No voiceprints are stored.

  • Four sections, every meeting

    Key points, decisions, questions and next steps, drafted on your server.

  • Filed by matter

    Each meeting lives under its client matter, with per-matter access.

For your security team

Component What it is
Deployment Docker on one Linux server, in your office or own cloud account. No GPU required. Air-gapped.
Outbound traffic None. No calls to AI companies, to us, or for telemetry. License checked offline.
Speech to text Parakeet-TDT 0.6B v3
Speaker labels pyannote segmentation 3.0 and TitaNet-Large, within a single meeting only
Summaries Gemma 4 open-weights model, run with llama.cpp
Storage PostgreSQL 18 on an encrypted volume, with a per-firm key
Transport TLS between the laptop, the server and your attorneys
Capture app Signed app for macOS and Windows

Reflects the current build. Model versions may change as better open models are released.

Product roadmap

Working today

  • Zoom meetings
  • Named transcripts
  • Key points, decisions, questions and next steps
  • Local recordings
  • Notes filed by matter

Next

  • Calendar sync: start notes in one click
  • Consent prompt before recording
  • Search across meetings
  • Retention and legal holds

Later

  • Suggests notes as a call starts
  • Microsoft Teams integration
  • Google Meet integration
  • Filing into iManage or NetDocuments

Order and timing are shaped with our design-partner firms.

What a clean record unlocks

Search every meeting by matter

“What did the client actually say in March?” takes seconds, not digging.

Memory that survives turnover

Matter context stays with the firm when people move on.

Data for your private AI

Clean, matter-linked notes inside your walls, ready for whatever AI your firm adopts next.

Keeping privileged conversations in-house is the reason to adopt. This is the upside.

What an evaluation looks like

  1. Step 1

    Security review

    Your IT and security teams review the architecture and test it on their own terms.

  2. Step 2

    Pilot

    A practice group uses it on real meetings, on a server you control.

  3. Step 3

    Your decision

    Your counsel and risk team reach their own conclusions.

Early firms help shape the product around their workflows as design partners.

We’ve shown you the build.
Your counsel draws the conclusions.

Every claim on this page is about where data flows and where it can’t.
The privilege question belongs to your firm.

Request a demo