Dder

Privacy policy

Dooder has no accounts and no analytics, and with your own AI there's no Dooder server in between. This page explains where your information does go, so you can decide whether you're comfortable with it.

Last updated 12 November 2026. Dooder is made by Stuart Breagan, an individual based in Sydney, Australia ("I", "me"). Questions: support@dooder.ai.

The short version

  • I don't keep anything you say to Dooder, anything it answers, or anything about you.
  • Dooder runs on your Apple Watch and iPhone. Your voice becomes text on your Watch and the audio never leaves it. The exception is Apple's dictation, if Dooder uses it, which Apple handles.
  • With your own AI, your questions go from your Watch straight to the AI provider you chose, and to apps only if you connected them.
  • With Dooder AI, your questions go through Dooder's relay to Anthropic to be answered. Dooder doesn't store your conversations on its servers.
  • Your AI key or sign-in is stored in your Watch's secure storage, not on your iPhone and not with me.

What stays on your devices

These are stored only on your Watch and iPhone, and are removed if you delete Dooder:

  • Your recent conversations with Dooder and the cards it showed you.
  • Things Dooder remembers about you, which you can see and delete in the iPhone app under Memory.
  • Your settings, saved prompts, connected app details and Dooder's look and voice.
  • Your Claude API key or ChatGPT sign-in, in the Watch's secure storage (Keychain), on that device only. Your iPhone holds it just until your Watch has it, then deletes its copy.

What Dooder keeps on your iPhone, such as past chats and memory, is part of your iPhone's backups like other app data. Keys and sign-ins aren't backed up.

What goes to your AI provider

To answer you, your Watch sends a request directly to the AI provider you chose (Anthropic for Claude, OpenAI for ChatGPT). It contains:

  • The words of your question and the conversation so far.
  • What Dooder remembers about you, so answers can take it into account.
  • Only when a question needs it: your approximate location (from Apple's location services, rounded to about a kilometre), events from your calendar, or your reminders.
  • Results from connected apps when Dooder looks something up there for you.

Your provider handles this under its own terms and privacy policy, through your own account. Anthropic and OpenAI each keep API requests for up to 30 days by default, longer when needed for safety or legal reasons. Web searches are done by your AI provider. I'm not a party to any of it and never receive a copy.

Dooder AI and invite codes

Dooder AI is for people without their own AI account. It works like Claude above, except that your Watch sends each request to Dooder's relay (a small service I run on Cloudflare), which passes it on to Anthropic using my account and sends the answer back.

  • What the relay doesn't store: your questions, Dooder's answers, what it remembers about you, or your IP address. It handles each request only while forwarding it, and its logs record only whether a request worked (a status code), never what was in it.
  • What it keeps for each invite: a name (your Reddit name or email), the email the code was sent to, a public key made by your Watch (the private half never leaves it), your time zone for the weekly reset, and how much of your allowance you've used, with daily totals of requests and tokens. These are kept while your invite is active, and deleted when you ask.
  • Wake-ups: when an app is still making something after you leave Dooder (a Canva design, say), your Watch gives the relay its push token and when to wake it, so Dooder can check and tell you when it's ready. The relay never learns what's being made, and the token is deleted once it's done (a scrambled copy is kept for an hour for rate limits). With your own AI, this isn't used.
  • Anthropic handles the requests under its commercial terms, which don't allow training on them, and keeps API requests for up to 30 days by default.
  • Cloudflare runs the relay and handles the connection under its own privacy policy.

Asking for a code. The form at dooder.ai/beta sends what you type (email, optional Reddit name, Watch model, what you'd use Dooder for, and whether you have an AI account) to the relay, where I read it. Cloudflare Turnstile checks that you're a person. For rate limits, the relay keeps a scrambled version of your IP address that can't be reversed and changes every day, for an hour. Emails are sent through Resend. I keep requests until the beta ends, or until you ask me to delete yours.

Connected apps

When you connect an app (for example Notion, Todoist or Canva), you sign in on that app's own page. The sign-in is given to your Watch, which then talks to the app directly. Dooder reads from an app when you ask, and asks for your OK on the Watch before changing anything there.

Downloads and look-ups that aren't your AI

  • Apple's dictation. Dooder normally hears you on the Watch. Until its speech model is ready, or if the microphone doesn't start, it uses Apple's dictation, which Apple processes under its own privacy policy.

  • Speech model and voice. During first-time setup your iPhone downloads Dooder's speech model (NVIDIA Parakeet) and Dooder's voice (Paradee), both published by FluidInference, once from Hugging Face. Hugging Face sees an ordinary download request from your iPhone.

  • App library. When you browse apps to connect, your iPhone searches the official MCP Registry (registry.modelcontextprotocol.io) and your iPhone and Watch load app icons from the apps' own websites.
  • Maps. Directions use Apple Maps. Opening a place on your iPhone uses Google Maps or Apple Maps, under their own terms.

None of these requests include your questions, your conversations or anything Dooder remembers. The only pictures Dooder loads are previews of what a connected app made for you (a Canva design, say), and only from that app's own site. The AI can't add pictures or links to a card, so a card can't make your Watch contact a site you didn't connect.

What I collect

With your own AI, nothing. Dooder has no analytics, no tracking, no advertising and no crash reporting of its own. With Dooder AI or an invite request, only what's listed under Dooder AI. If you choose to send feedback or a crash report through TestFlight, Apple shares it with me, along with the device and app details Apple includes. If you email me, I'll have your email.

This website

dooder.ai sets no cookies and runs no analytics. It's hosted by Vercel, which keeps standard server logs (such as IP addresses) for security and reliability under its own privacy policy.

Your choices

  • Delete what Dooder remembers, or your past chats, in the iPhone app under Memory.
  • Remove your AI key or sign out under AI provider. You can also revoke a Claude key in the Claude Console.
  • Disconnect any connected app under Connected apps.
  • Delete Dooder from your Watch and iPhone to remove everything it stored.
  • To have your invite or invite request deleted, email support@dooder.ai.

Children

Dooder isn't intended for children under 16.

Changes

If this policy changes, the new version will be posted here with a new date.

Contact

Stuart Breagan, Sydney, Australia. support@dooder.ai