Cohearo · Security & privacy

What leaves your device, and exactly when

Most transcription tools have to take your audio to work. Cohearo's architecture means it never needs to. This page is the specific version of that claim — what is processed where, what we store, for how long, and what we don't yet certify.

Never leaves

Your audio

Recording, transcription and speaker separation all run on your processor. There is no code path that uploads an audio file — not for processing, not for quality, not for support.

  • Captured, transcribed and diarized locally
  • Held on your disk for 72 hours so you can export it
  • Removed automatically after that window
  • Works with the network off
Leaves when you ask

Transcript text, for AI

Turning a raw transcript into minutes or a user story runs on our servers. That step sends the transcript text — never the audio — and only for the recording you pressed the button on.

  • Sent only on an explicit enhancement request
  • Text only: no audio, no voice data
  • Processed and returned, not used to train models
  • Skip it entirely and the app still does its main job
Leaves if you choose

Transcripts you sync

Whether a transcript reaches our servers depends on the mode you pick at sign-in. Cloud mode syncs new transcripts automatically; On-device mode keeps them private unless you push one yourself.

  • Chosen once at sign-in: On-device or Cloud
  • On-device recordings stay private by default
  • You can still push a single recording to the cloud yourself
  • Only synced transcripts are reachable by an assistant over MCP
Data inventory

Every kind of data, and where it sits

If your security reviewer wants one table, this is it.

Meeting audio

Where it lives
Your disk only
Retention
72 hours, then removed
Who can reach it
You. It is never transmitted.

Transcript (local)

Where it lives
Your disk
Retention
Until you delete it
Who can reach it
You

Transcript (synced)

Where it lives
Our storage, encrypted at rest
Retention
Free plan: up to 7 days · Paid plan: until you delete it or close your account
Who can reach it
You, and assistants you connect over MCP

Text sent for AI enhancement

Where it lives
Our servers, in transit
Retention
Processed and returned; not retained for training
Who can reach it
The model provider processing the request

Speaker labels

Where it lives
With the transcript
Retention
Same as the transcript
Who can reach it
Same as the transcript

Speaker voice profile

Where it lives
Your disk; the voice embedding never leaves it
Retention
Until you delete the profile
Who can reach it
You. The voice embedding is never transmitted; in Cloud mode, the profile's identifier and name may sync so the same person is recognized across devices.

Account and billing

Where it lives
Our systems and our payment processor
Retention
While the account exists, plus statutory records
Who can reach it
You and our support team

Product analytics

Where it lives
Our systems
Retention
Aggregate usage counts
Who can reach it
Us. No transcript content is included.
Your controls

What you can do about it

All of these work from inside the app, without contacting us.

Airplane mode

Still recording

Work fully offline

Disconnect entirely and everything that matters still runs. This is the honest test of the claim.

Export audio

Within the 72h window

Take it with you

Save a recording's audio to your device any time inside its 72-hour local window. After that it's gone, on the same clock that clears it automatically.

On-device

Push to cloud anytime

Choose your sync mode

Pick On-device or Cloud when you sign in. On-device recordings stay private by default; push a single one to the cloud later if you want it there.

MCP client

Label & rename only

Revoke an assistant

Disconnect an MCP client from your dashboard at any time. Access is read-mostly — it can label and rename, never upload, record or delete.

Being straight about it

What we don't claim

A security page that only lists strengths isn't much use to the person who has to sign off on it. So, plainly: we are a small team, and we do not hold a SOC 2 or ISO 27001 certification today. We are not going to imply otherwise with a badge.

What we offer instead is an architecture that makes most of those questions moot — audio that never transmits cannot be breached on our side — plus a direct line to the people who built it. If your review needs answers we haven't published, ask and we'll write them down.

Test the claim yourself

Install it, pull the network cable, and record a meeting. That is the whole demonstration.

The full policy

Security and data handling, in full

Version 0.2.0

Download as PDF