Memory Notch

AMBIENT SCRIBE

What "secure" should mean for an ambient scribe

An ambient scribe listens to a conversation and writes it up, with nobody dictating to it. The word "secure" is attached to almost all of them. It is worth knowing what it is being used to mean.

Three different things vendors call secure

Encrypted in transit and at rest. The audio travels over TLS and is stored encrypted. This is table stakes and says nothing about who can read it at the other end. Every serious vendor does this.

Contractually protected. The vendor signs an agreement about what it will and will not do with your data: a BAA under HIPAA, a DPA under GDPR, a commitment not to train models on your recordings. This is real protection, and it is the right answer for regulated work. It is also a promise rather than a property. It holds as long as the company does.

Never transmitted. The audio is processed where it was captured and no copy is sent anywhere. There is no retention policy to read because there is nothing retained, and no breach of the vendor can expose a recording the vendor never had.

All three are legitimate. They are not the same, and a page that says "secure" without saying which one it means is not telling you much.

Questions worth asking

The clinical case is its own thing

If you are documenting patient encounters, the decision is not really about architecture. It is about whether a vendor will sign a BAA and stand behind it, and whether the note lands in your EHR correctly. Abridge, Nuance DAX and the clinical scribes are built for exactly that, and they are good at it.

Memory Notch is not a clinical product. It has no BAA, no EHR integration and no claim to HIPAA compliance. Do not use it to document patient care.

Where an on-device scribe fits

For the conversations that are confidential without being regulated: a client call about money, a negotiation, a founder conversation, an interview, a consultation with someone who assumed the room was the room. In those, the strongest version of secure is the one where the recording was never transmitted in the first place.

Memory Notch transcribes and separates speakers on your Mac's Neural Engine, writes the note into ~/memory as plain Markdown, and never uploads a recording. See how it compares or try the demo.