Privacy

Effective 2026-10-19

cue is operated by HumanMeetsAI, a trading name of DEYI MEDICAL CONSULTING PTE. LTD., a company incorporated in Singapore — the data controller for everything this page describes. This page says what we hold about you, why we hold it, who else touches it, how long it stays, and how to get rid of it. It covers cue.humanmeetsai.com and the free cue apps for iOS and Android.

The short version: we hold what you gave us and what the product had to record to work. There is no third-party analytics SDK, no advertising, and we do not sell or rent anything about you to anyone.

What we collect

• Your account
  Your email address — that is the identity cue signs you in as — plus an optional display name and avatar, records of a Google or Apple sign-in you linked, your active sessions, and the one-time sign-in codes we email you.
• What you make
  Your decks and documents and their source, files and images you upload, the revision history of every deck, the chat messages you exchange with the authoring agent, notes and feedback people leave on a deck, and change proposals sent between forks.
• Sharing
  The email addresses you share a deck with, the share groups you define, and the ids of secret share links you create. Your contacts list is derived from who you have shared with and who has attended your presentations — we do not import or read an address book.
• Read receipts
  For decks shared with tracking on, we record who opened them, how far they got and how long they stayed. A signed-in reader is recorded by email; anyone else by a first-party id in a cue_v cookie. If your browser sends Do Not Track or a Global Privacy Control signal we record an anonymous open and set no cookie.
• Product analytics
  First-party only, on our own servers, with no third-party analytics SDK anywhere in the product. Visitor identifiers are HMAC-hashed before they are stored, and Do Not Track is honoured.
• How you found us
  If you answer the optional “How did you hear about cue?” card on your home page — one choice, plus a short note if you pick “somewhere else” — we store that answer on your account and delete it when you delete your account. Skipping it is stored the same way, so we never ask twice.
• Live presentations
  Who attended: the email of a signed-in viewer, and any email a viewer types into the wrap-up form at the end. See the section below for audio, video and captions.
• Screens
  For each display you pair as a cast target: the name you gave it, and its last IP address and browser user agent — that is how you tell two screens apart and how we can tell a stolen pairing code from a real one.
• Billing
  If you buy an author plan, Stripe handles the payment and holds your card details. We never see or store a card number. We keep your Stripe customer and subscription ids so we know what you are entitled to. When you delete your account we cancel every live subscription and ask Stripe to delete the customer record itself, so the card details it held for you go with it — Stripe keeps its own invoice and payment history, which it has to retain for tax and anti-fraud reasons.
• Trust and safety
  Reports made about a deck (including the reporter’s email if they were signed in), account holds, and the record of what we did and why.
• Security and operations
  Error records from the servers, and hashed rate-limit buckets that keep one caller from flooding an endpoint.

Why we hold it

To run the service you asked for: signing you in, storing and rendering what you author, delivering it to the people you chose, presenting live, telling you who read your work, and billing you if you bought a plan. Beyond that, to keep the service safe and working — abuse handling, rate limits, debugging — and to understand, in aggregate, which parts of the product people actually use. We do not use your content to train models, and we do not sell it.

Live rooms, audio, video and captions

When you present live, your microphone and camera are relayed through our own media server to the people in the room. The live media relay is not recorded. Separately, a presenter who has given written consent may have their caption microphone audio recorded for the quality evaluation described below. We do not record audience microphones or caption video.

Captions are different, and we would rather be plain about it. On the web, your microphone audio is streamed to Microsoft Azure Speech for recognition and translation using a short-lived token. Azure-enabled versions of the iOS and Android apps use the same service. Older iOS versions use Apple’s speech framework, which may send audio to Apple’s servers; older Android versions use the device’s recognition service. Final caption segments — the transcript in the language you spoke — are stored on our servers against the deck and the session so that you can read back and export what was said. When you or someone you shared a talk with reads it in another caption language, we cache that machine translation against the transcript rather than re-translating it on every read; the cached translation is deleted along with the transcript. The deck’s owner can delete a session’s transcript at any time from the deck’s captions surface, and transcripts are deleted with the deck and with the account.

Every presentation also leaves a talk record on the deck: when it started and ended, how long it ran, how many people joined, and how many caption lines were captured — counts, never who. The deck’s owner keeps that record. Nothing beyond it is visible to anyone else unless the presenter chooses to share it: a talk record can be shared, as a transcript and, if the presenter asks for one, a short AI-written recap, either with just the people who joined that talk or with everyone who can open the deck. The presenter can ask for a recap of any talk that has a transcript; it is generated from the transcript text only — never from audio — stored per language against the talk, labeled “AI-written” wherever it is shown, deleted along with the transcript, and never used to train a model. Sharing is always an explicit, per-talk choice, off by default, and a share expires 90 days after it is made unless the presenter renews it. Anyone in the room is told about this on their own screen the moment transcription starts — nothing is captured before that — and told again on the screen that closes the talk.

Caption quality recording is off by default. With prior written consent, we keep the presenter’s microphone audio, intermediate and final captions, translation inputs and outputs, and the relevant deck and slide context for evaluation. The consent names the vendors and routes under test and covers evaluation only, with no provider training or retention. Raw capture is kept for 30 days. Explicitly promoted evaluation clips and segments remain until consent is withdrawn. Named operators access recordings through audited tools; annotators receive only promoted clips under a confidentiality agreement. The presenter can download and delete their capture. Withdrawing consent or deleting the deck or account removes raw capture, promoted material and associated flags.

Any viewer can flag a caption from the caption line. We keep only that segment’s source and displayed text, language, slide and room, plus an irreversible participant hash to limit repeated reports. Flags are kept for 90 days unless explicitly promoted into a consented evaluation set. Caption text and evaluation context are excluded from application logs and Langfuse traces; language-model caption calls record only operational metadata such as model, latency and token counts.

AI processing

When you author with the built-in agent, your instructions and the deck’s source are sent to a large language model to produce the change you asked for. Today that model is served by Anthropic (or an Anthropic-compatible gateway) or by Microsoft Azure AI Foundry. We keep traces of those requests — including the prompt and the completion — on our own self-hosted Langfuse instance, so that a bad generation can be diagnosed. Nothing you author is used to train a model.

Caption translation and automatic deck glossaries may use language-model providers, including Anthropic, reached through OpenRouter; we send the caption text and deck context needed for those functions.

Content an AI produced is still your content, and it can be wrong. Review it before you publish it.

The same applies to a talk’s recap: it uses the same model provider as authoring, under the same trace posture above, is generated text, always labeled “AI-written” wherever it is shown, and — like every other AI output on cue — is never used to train a model.

Who else processes it

• Microsoft Azure — hosting, database, and the speech recognition and translation behind live captions.
• Stripe — payments and subscriptions.
• Resend — sending email (sign-in codes, share notices, moderation notices).
• Google and Apple — if you sign in with them.
• Cloudflare — the CDN that serves published decks and documents.
• Anthropic (or an Anthropic-compatible gateway) and Microsoft Azure AI Foundry — the language models behind authoring.
• Langfuse — LLM tracing, self-hosted by us on our own infrastructure.

These providers process data on our instructions to deliver the service. Some of them operate outside your country; where your data crosses a border we rely on the provider’s standard contractual protections.

Cookies

cue sets first-party cookies only, and only these:

• authjs.session-token / __Secure-authjs.session-token — keeps you signed in.
• cue_v (1 year) — the anonymous reader id behind read receipts. Not set if you send Do Not Track or Global Privacy Control.
• cue_ref (30 days) — remembers which link brought you here, so we can tell what is working.
• cue_lv (180 days) — which of two landing-page wordings you saw, so the page reads the same every time you come back. Not set if you send Do Not Track or Global Privacy Control.
• cue_follow_intent — carries a “follow this author” click through the sign-in you were sent to.
• cue_admin — the operator console session. Only operators ever get one.

How long we keep it

• Read receipts: 180 days, then deleted.
• Server error records: 90 days, then deleted.
• Talk records: until the deck or the account goes. Transcripts, their cached translations, and recaps: until you delete them, or until the deck or the account goes. Talk shares: 90 days, unless renewed.
• Everything else: until you delete it, or until you delete your account.

Your choices and rights

You can see and change your account details in the app. You can delete a deck, a document, an uploaded file, a share, or a caption transcript yourself, at any time. You can turn read-receipt tracking off for a deck when you share it.

You can delete your whole account — on the web at /account → Delete account, or in the iOS app at Settings → Delete account. That deletes the decks and documents you own and their published artifacts, your files, your sharing records, your read receipts, your sessions and linked sign-ins, and cancels any subscription and deletes your Stripe customer record. It is irreversible and we cannot undo it for you. Billing comes first: if we cannot reach Stripe to cancel your subscription we delete nothing at all and ask you to try again, rather than leave you paying for an account that no longer exists.

Decks kept on your phone. In the iOS and Android apps, decks you keep offline — and, because the “keep my recent decks offline” setting is on by default, decks you own and the last ten you opened — are downloaded onto your device in the background, over Wi‑Fi (or on any connection when a deck is under 5 MB); on Android that happens while the app is open rather than in the background. You can turn that off, change the storage limit, and remove every stored copy at Settings → Offline. Stored decks are protected by the device’s own data protection — iOS file protection, Android app‑private encrypted storage — and are excluded from device backups; we hold no key to them and they are never uploaded anywhere. Two consequences worth saying plainly: the app re-checks each stored deck with us whenever it is online, and a deck you are no longer allowed to see is deleted from the device at that moment — but a share or a grant revoked while the device is offline only takes effect the next time the app reaches us, and a copy already on a device cannot be recalled, exactly as with any download.

For a copy of your data, a correction, a restriction on how we use it, or an objection to it, write to info@humanmeetsai.com from the address on the account and we will answer. Depending on where you live you may also have the right to complain to a data protection authority, and — if you are in California — to know what we collect and to ask that it not be sold or shared. We do not sell or share personal information, so there is nothing to opt out of, but the request is still yours to make.

What survives deletion

Two things, deliberately. Records of abuse — reports, holds and the moderation actions we took — are kept as a minimal record after an account is deleted, because otherwise deleting an account would erase the history that stops the same abuse being repeated. And the do-not-email suppression list survives, because the only way to honour “never email me again” is to remember it. Both are kept on the basis of our legitimate interest in keeping the service safe and in honouring your own opt-out.

Backups and CDN caches are purged in the normal course rather than instantly. And once a deck has been public, we cannot recall copies other people already downloaded — no platform can.

Children

cue is not directed to children under 13, and we do not knowingly collect anything from them. If you are older than that but not yet old enough to consent to this on your own where you live, a parent or guardian has to agree for you. If you believe a child has an account, write to info@humanmeetsai.com and we will remove it.

Security

Everything travels over TLS. Sign-in codes, visitor identifiers and rate-limit keys are hashed rather than stored raw, secrets live in the environment and never in the source, and published decks are served from a separate origin so a deck’s own code can never reach your session. No system is perfect; if you find a hole, please tell us at info@humanmeetsai.com before you tell anyone else.

Changes

The product changes often and this page changes with it. The effective date at the top tells you when it last did. If a change materially affects what we hold or why, we will say so by email to account holders rather than leaving you to notice.

Contact

Privacy questions and data requests: info@humanmeetsai.com. Abuse and takedown: an@humanmeetsai.com. See also our terms and acceptable-use policy.

← Back to cue