AI and Third-Party Services Notice

Version 1.0. Effective 28 October 2026.

This notice explains which parts of Dictate Safe and of our website use artificial intelligence (AI), where each part runs, what changes if you choose a cloud provider, what we receive, and which other companies take part in the Services. It adds to section 5 of the Privacy Policy, section 6 of the Terms of Service and sections 8 and 9 of the Licence Agreement.

1. The short version

QuestionAnswer
What uses AI?In the App: speech recognition, optional text cleanup, and the Assistant (Beta). On our website: an AI agent in a chat window and on a voice call (section 2.4).
Where does it run?In the App, on your computer, using open-source models that the App downloads when you ask. The website's AI agent runs at AI and voice providers that act for us (section 2.4).
What changes if I choose a cloud provider?The App sends data straight from your computer to that provider, using your own key. Section 4 says what.
Do you receive my voice, text or chats?Not from the App. If you write or talk to the AI agent on our website, we receive what you write or say there (section 5).
Do you train AI on my dictation or chats?No. We do not receive your dictation or Assistant chats, so we have nothing to train on, and we do not use website chats or calls to train AI models.
Can the AI be wrong?Yes. Check what it produces before you rely on it.
Who decides what the Assistant changes?You do. Larger changes wait for your approval, and smaller ones can be undone.
Where do I report a problem?support@dictatesafe.com, or the Contact page.

2. Which parts of Dictate Safe use AI

2.1 Speech recognition turns your voice into text. It is on whenever you dictate.

2.2 Text cleanup is optional and is off until you turn it on. It tidies the finished text.

2.3 The Assistant is a Beta feature and is off until you turn it on. It is a chat assistant that helps you manage your tasks, goals and reminders on your computer. It answers in English only.

2.4 The AI agent on our website is a chat window and, where a microphone button is shown, a voice call. It answers questions about Dictate Safe, gives prices from our price list, can email a link or a one-time code, and can give a discount code from a short list we set. It is not part of the App and does not run on your computer. AI providers that act for us write its replies: at any time one or more of Anthropic, OpenAI, Google (Gemini) and Groq, chosen in our own settings. Retell runs the voice call. It can be wrong. A price it quotes is from our price list at that moment, and the pricing page and the checkout decide what you pay. It cannot approve a refund or change your account, and a person on our team can read and take over any conversation. The Privacy Policy (section 5.4) says what is stored and for how long.

3. Where each part runs by default

By default the three parts run on your computer. The AI models they use are open-source. The App does not include them. It downloads them from third-party hosting services when you press a download button, and those services see your IP address and the request. Their licences are listed in the App, under Settings, About, Open-source licences. Your recordings, text, History and chats stay on your computer unless you choose a cloud provider.

4. What changes when you choose a cloud provider

You can switch on a cloud provider by saving your own key and choosing the provider in the App. We do not provide keys, accounts or servers for this. The data does not pass through us. The App sends it straight from your computer to the provider, under your agreement with the provider.

What you switch onWhat goes to the providerProviders the App supports
Cloud speechYour recording and your vocabulary listGroq (privacy policy)
Cloud cleanupThe finished text (only the parts that need work, plus the neighbouring sentences, or the whole of a short text) and your vocabulary listGroq, OpenAI (privacy policy), Anthropic (privacy policy)
Assistant on a cloud engineYour messages and the conversation so far. In the default mode it also sends task titles, task notes, goals and reminders. Files and prompt text go only after you press "Send this once" on a card that shows exactly what would be sentGroq, OpenAI, Anthropic

4.1 What never goes. Your History, Scratchpad, dictation notes and snippets list are never sent to a cloud provider. A snippet that you speak is expanded before cleanup, so its expanded text can be part of the text that a cleanup provider receives.

4.2 "Send this once". For a file or a saved prompt, the App shows a card with what would be sent. Nothing goes until you press "Send this once" on that card. The card asks every time.

4.3 What the provider does with it. A cloud provider may keep or use what it receives, as its own terms say. We do not control that, and we cannot delete the data for you. Read the provider's privacy policy before you switch it on. The links are in the table.

4.4 Switching back. You can go back to the on-device models in the App's settings, remove your key, or turn on Offline Lock. Offline Lock stops every connection that the App makes itself, including to cloud providers. It does not stop buttons that open your web browser.

5. What we receive

The AI agent on our website is the one exception. If you write or talk to it, we receive and keep what you write or say there, and our AI and voice providers receive it to answer you, as the Privacy Policy (section 5.4) explains. We do not keep the audio of a call.

We receive none of the content you dictate or write in the App. That covers audio, transcripts, History, notes, snippets, vocabulary, prompts, tasks, goals, Assistant chats and files you give the Assistant. Section 3.4 of the Privacy Policy lists what we do not collect, and section 6 lists what the App does send to our servers. If you choose to send us a message, a file or a screenshot, we treat it as a message under the Privacy Policy.

6. Training

We do not use your dictation, History, notes, tasks or Assistant chats to train AI models. We could not, because we do not receive them. We also do not use the chats and calls that people have with the AI agent on our website to train AI models. Our team reads them to answer, to keep the service safe and to improve the agent's instructions.

7. Telling you that the Assistant is AI

Article 50(1) of the EU Artificial Intelligence Act says that people who talk directly with an AI system have to be told that it is one. The Assistant is an AI system. The first time you open it, it tells you that it is software and not a person, and it tells you whether it is running on your computer or with a cloud provider you chose. Speech recognition and text cleanup do not hold a conversation with you. The AI agent on our website tells you in its first words, in every chat and every call, that it is an AI agent and not a person, and says so again if you ask.

8. Accuracy and limits

8.1 Speech recognition can mishear. It can add, drop or change words. The speech models that the App downloads are English models.

8.2 The Assistant can be wrong. It answers in English only. It is not a professional adviser and does not give medical, legal or financial advice.

8.3 Check text before you send, publish or rely on it. Do not use the App as the only safeguard where a mistake could cause injury or serious harm. Section 6 of the Terms of Service and section 8 of the Licence Agreement say more.

9. Your control over what the Assistant does

9.1 The Assistant changes only your tasks, goals and reminders on your computer.

9.2 In its default mode the Assistant applies smaller changes at once and shows you what it did, with an Undo button for a limited time. For larger changes, such as removing tasks or changing many tasks together, it shows a card and waits. Nothing happens until you press the button on the card. You can stop the Assistant while it works.

9.3 With a cloud engine, files and prompt text go to the provider only after you approve them, as section 4.2 says.

10. Help in a crisis

The Assistant is not a counsellor and cannot give you the help you deserve. If you tell it that you are in crisis, it shows a help card. The words on the card are fixed text that we wrote, not an AI answer. The card shows helpline numbers for your region where it has them (India, the United States, Canada, the United Kingdom, Ireland and Australia), and a button that opens a worldwide directory of helplines in your browser. If you are in danger now, call your local emergency number.

11. Reporting a problem or a harmful output

If the Assistant, speech recognition or cleanup gives an answer that is harmful, unfair or plainly wrong, write to support@dictatesafe.com or use the Contact page at https://dictatesafe.com/contact. Tell us what happened and when. Please send the text or chat only if we need it to understand the problem, because anything you send us reaches us as a message. If a cloud provider produced the output, you can report it to the provider as well. To report a security weakness, use security@dictatesafe.com instead.

12. Other companies that take part in the Services

Each company below acts for us or on its own account. The sub-processors page says which, what each receives and where.

  • Amazon Web Services hosts the Website and its database.
  • Resend sends our email: service emails, news, offers and product emails.
  • Anthropic, OpenAI, Google (Gemini) and Groq write the replies of the AI agent on our website and in Support, one or more at any time.
  • Retell runs the voice call with the AI agent.
  • Razorpay takes payments and refunds.
  • Google provides sign-in, Google Analytics and Google advertising features, and its workspace tools for our team's email.
  • GitHub provides sign-in.
  • Microsoft provides Clarity, which records how people use our pages, and the Microsoft Store for people who install from it.
  • Meta provides the Meta Pixel and the places where we advertise on Facebook and Instagram.
  • AppSumo sells licences on its marketplace for people who buy through it.
  • Hosting services serve the AI model files that the App downloads.

Analytics, Clarity, the Meta Pixel and Google advertising features load according to your region. In the EU, the EEA, the UK, Switzerland and every other place except India and the United States, they load only after you allow them in the cookie box. In India and the United States they load by default and stop when you press "Necessary only" or your browser sends the Global Privacy Control signal. The Cookie Policy explains.

13. Contact and changes

Questions: privacy@dictatesafe.com. We may update this notice. When a change matters, we will tell you by email or in the App, and we will update the version and date above.

Document history

VersionDateWhat changed
1.028 October 2026First version in this form.