AI Disclosure

Equerry uses large language models (LLMs) to make your personal content searchable, answer your questions about it, and run your Routines. This page is the plain-language version of how that works — what gets sent where, what is and isn't retained, and the controls you have.

What we use AI for

• Understanding what you save. When you forward an email, save a note, or record a voice memo, we use an LLM to summarize it, classify it, and extract structured facts (for example: "Jen is your wife"; "the Acme Q3 review is on Friday").
• Answering your questions. When you ask Equerry something, we retrieve the most relevant excerpts from your memory and send them to an LLM, which composes the answer with citations back to the original source.
• Running your Routines. Each Routine — Morning Brief, Watcher, Reminder, Meeting Prep, Research, or a custom one — uses an LLM to compose its output from your memory and any external sources you've authorized.
• Voice transcription. Voice memos are transcribed using a speech-to-text model. The transcript is what becomes searchable; the original audio stays in your iCloud.
• Image and PDF processing. Photos and PDFs are processed with optical character recognition and a vision-capable LLM to extract text and structured information. The original files stay in your iCloud.

What the LLMs see

LLMs receive only what is necessary for the specific task in front of them:

• The content you have actively saved or forwarded to Equerry — never your raw iCloud library.
• The question you have asked, plus the small set of memory excerpts our retrieval system identifies as relevant.
• Your assistant's persona settings, so its tone matches your preferences.
• If you have allowed calendar access for Meeting Prep or Morning Brief, the upcoming events in your calendar snapshot (titles, times, locations, and attendee names — never event notes or email addresses), only while composing that brief.

LLMs do not see your other notes, your conversations with other apps, your photo library, or anything else outside the explicit Equerry workflow.

Which vendors

Equerry routes work to the model best suited for the task at hand. The vendors we use today are:

• Anthropic (Claude family, hosted on Amazon Bedrock) — chat answers, reasoning, and extracting facts from what you save.
• Google (Gemini family) — understanding images and PDFs, and transcribing your voice notes.
• OpenAI — the text embeddings that make your memory searchable, a backup model when our primary is unavailable, and premium voice (later).
• Amazon Web Services (Bedrock) — the infrastructure that hosts and routes requests to the Anthropic models above.
• Exa — web search used by Routines and Asker mode. It receives your search query, not your stored content.

We may add or replace vendors over time as the model landscape evolves. When we do, this page is updated.

No training on your data

Every vendor in the list above operates under enterprise terms that prohibit using your data to train any model. This is non-negotiable on our end.

What vendors retain

Vendors retain inference logs (the input they received and the output they produced) for a short window — typically up to 30 days — for abuse detection and operational diagnostics. After that window, vendors delete the logs. Equerry separately keeps a debug log of vendor calls for 30 days, then deletes it automatically.

Where your raw content lives

The raw notes, voice recordings, photos, and PDFs you save to Equerry are stored in your iCloud account, on Apple's infrastructure, encrypted with your iCloud keys. Equerry's servers store derived representations — text excerpts, embeddings, extracted facts — needed to make the content searchable, but not the original files. A breach of Equerry's backend cannot expose your raw audio, photos, or PDFs.

Limits on what AI can do for you

Equerry's AI never takes consequential action on your behalf without an explicit confirmation. It will not:

• Send emails to other people on your behalf.
• Make purchases or move money.
• Schedule, accept, or decline calendar events without you confirming.
• Share your content with anyone outside the vendor list above.

AI can be wrong

LLMs can produce inaccurate, incomplete, or misleading output — even when answering questions about your own content. Equerry mitigates this by citing the original source on every answer, so you can tap through and verify. We strongly encourage you to verify before acting on anything material.

Your controls

• See what's been extracted. Settings → Your Memory shows every fact Equerry has derived from your content.
• Delete extracted facts. One tap, immediate.
• Delete source content. Removes the original note plus every derived fact, embedding, and excerpt.
• Pause ingestion. Turn off voice transcription, image processing, or email forwarding individually in Settings.
• Rate answers. The thumbs-up / thumbs-down on every chat response helps us improve the retrieval and prompting. We never share your rated content with vendors for training.

Changes to this disclosure

When we add or change vendors, change retention policies, or change the categories of work the AI performs, we update this page and notify you in the app.

Questions: support@getequerry.app.