Privacy
Where your meetings go, and who can read them.
A page for whoever has to approve Diurna at work: what stays on your computer, what goes through servers in Europe, who sees it and for how long.
Two ways of working
Only on your computer
Audio, transcript and minutes never leave the computer. The minutes are written without external AI models. No sync.
Cloud in Europe
Text goes to EU servers for AI minutes, Ask, wiki and sharing with your team. If you choose cloud transcription, audio passes through the EU server only to be transcribed and is not kept. You choose meeting by meeting.
The data path
-
Your computer
Records and transcribes. Audio and text encrypted, key in the system keychain.
-
Servers in the EU
Syncs only the meetings you choose, the wiki and reminders.
-
AI models
Minutes and Ask get text only. Audio goes only to cloud transcription, if you choose it. EU providers, zero retention.
-
Your team
Only the people you add, at the level you choose.
In Only on your computer mode, the path stops at the first step.
In detail
- Audio
- Recorded on your computer. It doesn’t leave your computer, except for cloud transcription if you choose it: then OVHcloud AI Endpoints in Gravelines, France, transcribes it and doesn’t keep it.
- At rest
- Audio and transcripts encrypted on your computer, with the key in the system keychain; on the server, data encrypted in the EU.
- In the call
- No bot and no extra participant. You record, from your own computer.
- Retention
- Audio deletes itself after the number of days you choose, or never. Transcript and minutes stay, so quotes keep working.
- Consent
- On the first recording of every series, Diurna asks you to inform participants and prepares the message to paste in the chat.
- Sharing
- Three levels: minutes, transcript, transcript with audio. Access is kept in the audit log.
- AI models
- Optional. No training on your meetings, no retention by the providers.
Usage statistics
To understand how the site and the apps are used we count visits and actions with PostHog, in its European cloud in Frankfurt. Meeting content never ends up in the statistics.
- On the website
- Pages viewed, button clicks, how far you scroll, where you came from (links and UTM parameters), browser and device. On public pages PostHog also records how you move around the page (session replay), with form fields always masked. On the sign-in and account pages and in the web app this recording is off.
- Cookies
- PostHog stores a random identifier in your browser, in a cookie and in localStorage, to recognise the same visit from one page to the next. If you sign in to Diurna, it links it to your account’s internal identifier, never to your email.
- IP address
- From the website PostHog receives your browser’s IP address and derives an approximate country and city. The Mac app does not send location.
- In the web app
- We only count which section you open (Meetings, Ask, Wiki…). Page addresses reach PostHog cleaned up: no meeting IDs, searches or sign-in link codes.
- In the Mac app
- Counts such as “recording started”, “notes generated” or “question asked”, with a random identifier not linked to your account. Never audio, text, titles, names, emails or file paths. Turn them off in the app’s Settings, under “Anonymous usage statistics”.
- Opt out
- The button below stops collection from this browser; your browser’s “Do Not Track” setting works too.
iPhone app
The iPhone app records in-person meetings with the phone’s microphone, or imports an audio file you pick. It always works with Cloud in Europe: audio goes through Diurna’s EU servers and then deletes itself.
- What it records
- The iPhone’s microphone, that is, the room. The iPhone doesn’t allow recording other apps’ calls, and Diurna doesn’t.
- Upload
- While you record, audio is uploaded in blocks, over an encrypted connection, to Diurna’s EU servers. Without a network it waits on the phone and leaves when the network is back. Each uploaded block is removed from the phone.
- Transcription
- From Diurna’s servers, audio goes to OVHcloud AI Endpoints in Gravelines, France, which transcribes it without keeping it. If you turn on transcription on the phone, the text is made on the iPhone and OVHcloud doesn’t receive the audio; the servers transcribe it only if the phone’s text is incomplete.
- Retention
- Audio deletes itself from the servers 24 hours after the meeting is ready. The cleanup runs every 6 hours, so it’s gone after 30 hours at most. Transcript, minutes and marked moments stay until you delete the meeting or the account.
- Consent
- Before the first recording, the app explains where audio and text go, lists the providers and asks you to always tell the people in the room, with a ready-made sentence to say. Without this confirmation the servers don’t accept the recording. For imports it asks you to use only audio you have the right to use.
- Marked moments
- The ★ you tap while recording (the moment and an optional note) go to the server with the recording: the AI minutes give those moments more weight.
- AI models
- For minutes, memory and Ask, the text takes the same path as on the computer: OpenRouter, to models with EU endpoints and zero retention.
- Deletion
- Deleting a meeting in the app also deletes its audio, transcript and marked moments from the servers. Deleting the account, from Account in the app, deletes everything. Account › I tuoi dati (the app is in Italian) explains where the data goes.
Transcription from the web page
From the page getdiurna.com/trascrivi anyone can upload an audio or video file, without an account, and receive by email a link to the transcript and the Summary. The file takes a separate path, shorter than the apps’ one.
- What we receive
- The file, the email to send the link to, the file’s name and length, and the chosen language. From the email and the IP address we also compute a fingerprint (HMAC with a secret key) that can’t be traced back to the addresses and is used for the daily limits. If we ask you to confirm the address, after you confirm the browser keeps a signed code, tied to the email’s fingerprint, so you don’t have to confirm again from that browser for a year.
- Where it goes
- The file reaches the Diurna server in Spain (Microsoft Azure) over an encrypted connection. OVHcloud AI Endpoints, in Gravelines, France, transcribes the audio without keeping it. The text goes through OpenRouter to models with EU endpoints (Azure OpenAI), with zero retention, which remove filler words and write the Summary. The email is sent by Resend, in Ireland. The upload status lives in the database in Germany (Supabase, Frankfurt).
- The link
- The email carries a link to a page on getdiurna.com where you read the transcript and the Summary, download them as Word or text, or delete them. The link contains a signed code that can’t be guessed, and anyone who has it can open the page: forward it only to people you choose. The page is kept out of search engines. The email carries only the link, never the text.
- Retention
- The audio is deleted as soon as the transcription is done. The email address is deleted right after sending, unless you asked to join the waitlist (see below). Text, Summary and file name stay on the server, for the link’s page, for up to 30 days from the upload; then they are deleted, and you can delete them sooner from the page. An upload that stalls for an hour, or isn’t finished within 8 hours, is deleted. Only the fingerprints, the file’s size and length and the outcome stay for 30 days, for the limits and to check for abuse; then the row is deleted too.
- Training
- Neither Diurna nor its providers use the files or the texts to train models.
- Consent
- Before sending, we ask you to confirm that you have the right to use the audio and that you have informed the people recorded. If you cancel the upload from the page, the file is deleted straight away.
- Waitlist (optional)
- If you tick “Let me know when the Diurna app is available”, once the link is sent your email address joins the waitlist for the app, with the language and the date. We only use it to email you when there’s a spot for you in the app, and we keep it for as long as you stay on the list. The email with the link reminds you and has a link to leave the list whenever you want. Without the tick, the address is deleted as above.
Google and Microsoft 365 calendars (optional)
Connecting a calendar is your choice: Diurna works without one. If you connect it, Diurna asks for read-only access and uses that data to show your events and match a recording to the right meeting.
- What it reads
- Title, date and time, location, organizer, attendees (name and email) and the video-call link for events between one day before and 14 days ahead. With Google the scope is calendar.readonly; with Microsoft 365 it is Calendars.Read. Diurna does not create, edit or delete events.
- What it is for
- Showing upcoming events in Today, offering to record at the right time, preparing the brief before the call and attributing the meeting to its attendees in the minutes.
- Where it stays
- Copied events stay on your computer, in the app’s local database. Access tokens sit in the encrypted store protected by the system keychain. Neither events nor tokens are sent to Diurna’s servers: only the title, date and attendees of the meetings you choose to sync do (see below).
- What we never do
- No selling or sharing with third parties for their own purposes, no advertising, no model training, no human reading of your events.
- How to revoke
- Settings › Calendars › disconnect the account: Diurna deletes tokens and copied events. You can also revoke access at myaccount.google.com/permissions or account.microsoft.com.
- If you sync a meeting
- Only for the meetings you choose to sync to the Cloud, Diurna stores on EU servers (application in Spain, database in Frankfurt) the title, date and time, video-call link and attendees (name and email) as part of the meeting. They stay until you delete the meeting or the account.
- AI models
- To write the minutes and answer “Ask”, the meeting text, the title and only the attendees’ names (never emails) go to a language model through OpenRouter, on EU endpoints (Azure OpenAI), with zero retention and no data collection: providers cannot store them or use them to train models. While a Google calendar is connected, the ⌘K bar’s smart suggestions, which use a different provider, are switched off.
- Data protection
- Encrypted connections (TLS); data encrypted at rest on your computer and on the servers; calendar tokens in the system keychain; each account sees only its own data; no human access to customer content except on explicit request for support. Deletion: disconnecting the calendar removes tokens and copied events; deleting a meeting or the account removes the server data.
Diurna’s use of data received from Google APIs complies with the Google API Services User Data Policy, including the Limited Use requirements: that data is not used for advertising, is not sold and is not used to train models.
Where the servers are
Application server in Spain (Microsoft Azure), database in Germany (Supabase, Frankfurt), cloud transcription in France (OVHcloud, Gravelines), service emails from Resend in Ireland, usage statistics with PostHog (EU cloud, Frankfurt). The full list of sub-processors comes with the DPA.
DPA
Data processing agreement for the Team and Enterprise plans.
Ask for it with your access request →A person, not a bot
Questions about data and privacy are answered by the founder, Francesco Del Sesto.
Write to me on LinkedIn →