Where your data goes.
Plainly, because you'll ask.
Where does my audio go?
Through your browser's Web Speech API — in Chrome, Google's speech service. Not to Hemi's server, which receives only text. No audio is retained.
Where do my transcripts go?
To Anthropic's Claude API, when you ask for a document. Nothing is sent before that.
What happens if I email a letter to a patient?
The letter and the address go to Resend, which delivers it. Resend sends mail for this part of the world from Tokyo and keeps its own records — recipient, subject, delivery outcome — in the United States. Email is not private in transit and the recipient's mail provider keeps a copy, which is why every send records that the recipient agreed to receive it this way. Hemi stores a checksum of exactly what was sent. Emailing is off for every account until an administrator turns it on.
Is my data used to train AI models?
No. Content is processed by Anthropic's Claude API to generate structured notes; Anthropic does not use API traffic for model training.
Is it compliant?
No. It is a prototype on test data, and it needs audit logging and rate limiting on sign-in before it holds real records.
Does it connect to my EMR?
No connector exists yet — not an API, not an HL7 feed, not a file drop.
What if it mishears something?
Low-confidence phrases are flagged before you export, and gaps are marked rather than filled in. You are still the one signing it.
Can I edit what it produces?
All of it, freely. Each document is independently editable.
What does it cost?
Nothing yet, because it is not a product yet.
Which browser?
Chrome. Hemi uses the browser's built-in speech recognition, which is where that is best supported.
Is it bilingual?
Interface and dictation, yes. Formatted documents are English while the clinical vocabulary is still under review.