How your data is handled
This is a plain explainer, not a legal document. The binding versions are Privacy and Terms.
What Donna can see
What you say. Everything you type and everything you say on a call, plus the transcript of the call.
What you make. Your cards, tasks, notes, plans and comments, and any image or file you attach or share to Donna.
Your calendar. On iPhone, your phone's own calendars, once you allow it. On the web, your Google Calendar if you connect it — read and write.
What is connected for you. Channel and tool connections are not self-service yet. Where one has been set up for your account, Donna may read from it to answer you: WhatsApp and Telegram threads, and items in Linear, Notion or GitHub. Nothing is connected by default.
How you tend to work. Donna builds a short written description of your working patterns and uses it when planning your days. It exists to plan for you and goes nowhere else.
What Donna cannot see
Donna does not read your email inbox. Donna is not in iMessage. Donna does not know your location unless you ask a card for something that needs it, and never in the background. Donna does not see anything in an app you have not connected. And there is no Telegram bot — if you used one under the Donna name before, it is gone and has been for months.
What Donna stores, and where it lives
Most account data in Postgres is protected by per-user row-level security. We are still auditing every product path and do not claim complete per-user isolation across the whole service. Files and images sit in the same provider's storage. Traffic to us and to every AI provider is encrypted in transit.
Voice calls are not ephemeral. We keep the full transcript of each call, its summary and its duration, attached to your account.
We keep an audit log of changes and tool calls. One honest wrinkle: if you delete a card, the card goes but the audit-log line recording that it changed stays.
Where it goes
Donna is an AI assistant, so answering you means sending your words somewhere else. The full list is in the privacy policy; the short version is: ElevenLabs, Anthropic, Google, Z.ai, OpenRouter, OpenAI, Nous Research and Exa may each handle part of what you send, depending on what you asked for and which company is available. What goes with a request can include your name and email, your notes, lists, files and photos, what Donna remembers, and calendar events when you let her use your calendar — not just your question.
None of them is allowed to sell your data or pass it on for their own purposes.
We never say "your data stays with us", because it does not. It is the one sentence an AI assistant cannot honestly write.
Your voice, specifically
When you are on a call, your microphone audio goes from your device to Donna's server, and from there to OpenAI. OpenAI turns your speech into text, writes Donna's replies and speaks them.
When ElevenLabs is in use, it handles the call instead: ElevenLabs turns your speech into text and speaks Donna's replies, Anthropic writes the replies, and Google writes a short note about the call. All of them are named in the privacy policy.
So: your voice leaves your device, passes through Donna's server, and reaches the company running the call. That is the accurate description.
If you turn on Your voice, Donna also keeps a voiceprint so she can recognise you on your calls. It is optional, kept encrypted on Donna's server, not sent to any AI company, and deleted when you tap Forget my voice or delete your account.
Who can send on your behalf, and how the gate works
This is the part most people want to know, so here it is in full.
Donna can write a message to a real person. When Donna does:
- Nothing is transmitted when the draft is made. Writing it and sending it are two different acts, in the code as well as in the copy.
- The draft is read back to you word for word — not a summary of it, not "I've drafted something", the actual text.
- It sits there. You can edit it, skip it, or leave it. It stays a draft.
- It goes only when you say send. That confirmation is the one thing that authorises it, and it authorises exactly that one message.
Where a connected account has been set up for you, Donna can send on WhatsApp and Telegram after your approval. General self-service channel linking is still under review. Donna can draft for email and SMS, but sending on those channels is not live yet.
Your calendar is different: Donna changes it when you ask her to, without a separate approval step.
If you refuse a permission, Donna says so rather than working around it. Refuse the microphone and typing still works. Refuse the calendar and Donna tells you it cannot see your calendar instead of guessing.
What Donna remembers, and how to make it forget
Donna keeps one long-running conversation and remembers between sessions — that persistence is most of what makes Donna useful.
You are not asked to trust it blindly:
- Donna picks out what is worth remembering on her own, but sensitive things wait for your yes: health, money, legal matters, religion or politics, sexuality, and other people's private details.
- You can see every line Donna has kept, and Forget any of it. Forget is real: the line goes.
Forgetting removes it from Donna going forward. It cannot reach copies an AI provider already received.
How to delete everything
In the iPhone app: Settings › Account › Delete account. This deletes the account and its data, including what Donna's server keeps for you and your voiceprint, and signs you out. It does not immediately clear backups, logs, or copies already sent to AI providers.
Three things to know:
- Cancel your subscription separately. An active subscription is billed by Apple and lives in your Apple Account settings. Deleting here does not cancel it.
- It does not reach the AI providers. What they already received is theirs to retain under their own terms, which we have not verified.
- On the web there is no delete button yet. Until there is, email [email protected] from the address you signed in with and we will do it.
You can also ask for a copy of everything we hold, any time.
What we do not do
- No tracking. No tracking pixels, no advertising identifier, no tracking domains. The iPhone app declares this to Apple and the declaration is empty by design.
- No ad networks. None. Not now and not planned.
- No analytics SDKs. No PostHog, Segment, Mixpanel, Amplitude, Google Analytics — nothing of that kind is in the app. This website is delivered through Cloudflare, which counts its page views for us with Cloudflare Web Analytics. It uses no cookies.
- No cookie banner, because there is nothing to consent to beyond the cookie that keeps you signed in.
- No sale of your data, and no sharing with advertisers or data brokers.
- No profile built for anyone but you.
These are verifiable rather than aspirational: there is no such code in the app to find.
What we are still deciding
Honesty is the point of this page, so here is what is genuinely open.
How long we keep things. We have not set a retention window — not for our database, not for our logs, not for the audit trail. And we have not verified what each AI provider keeps, or for how long, or whether they use it to improve their models. So we make no promise in either direction. We would rather say this than publish a comforting number we cannot stand behind.
Whether consent is enforced in the release you are on. We record your consent to AI processing, but enforcement can vary by release. Do not assume every release blocks transmission until consent has been recorded. Withdrawing consent does not retrieve anything a provider has already received.
A formal security posture. We are a small, new product. We hold no security certification and have not been independently audited. Row-level security and encryption in transit are real; a SOC 2 report is not something we have. Treat genuinely sensitive material accordingly.
A delete button on the web. The iPhone app has one. The web does not yet.
Telling us about a problem
If you find a security problem, write to [email protected]. We would rather hear it from you than read about it. Tell us what you found and how to reproduce it, and give us a chance to fix it before you publish.