Privacy

Your speech is recognised on your own computer and is not sent anywhere. We do not listen to your conversations, we do not keep voice recordings, and we do not use what you say to train models.

Only what the work cannot happen without reaches our server: text to be formatted, the turns of a spoken conversation, and usage accounting. Below is exactly what, and why.

Speech recognition. The model runs entirely on your machine; microphone audio never leaves the device. That is why dictation also works with no internet.

The history of what you dictated. It is kept in your application folder and we cannot see it.

Conversation memory. What the assistant remembers about you from past conversations stays with you. It is passed to the server for the length of a conversation and forgotten together with it. One button in settings wipes it.

Keys and passwords. Your licence key is kept in the Windows Credential Manager. None of OUR model keys are on your computer - those live only with us. If you brought your OWN provider key, it sits in that same Credential Manager on your machine and never travels to us.

Text to be formatted. What you dictated is sent so that punctuation can be placed and slips removed. We do not keep it after processing.

Spoken conversation. Your voice travels through our server to the model and back. Through us, so that a provider key never ends up on your computer and so that usage is counted honestly. We do not keep recordings of the conversation.

Screen images - only while you have screen sharing switched on yourself, and only for as long as it is on.

Usage accounting: how much was processed, and when. Without the content.

So that one licence does not run on ten machines, we compute a fingerprint of your computer. It is a hash - an irreversible number from which neither serial numbers nor anything else about your hardware can be recovered.

Text and conversations are processed by third-party model providers: Google, Anthropic and OpenCode Zen. Only what is needed for an answer reaches them, and without your name or email. Your speech itself goes nowhere: it is recognised on your computer.

We pass data to nobody else and we do not sell it.

Usage accounting and your account - until you delete the account. Recordings of speech and conversations we do not keep at all.

You may take your data, correct it, or delete the account entirely. Conversation memory and history can be wiped inside the program without asking us.

Write to us: support@midri.app.

The Midri Voice service. Contact for anything: support@midri.app.

The legal details of the owner will be published here before we leave the beta and start taking payment. Today the service takes no money.