Privacy Policy

1. What we collect

1.1 Information you provide

• Account information — your email address, name and sign-in identifiers (including from Google or Apple sign-in if you use them).
• Optional profile answers — if you answer them when you first sign in or later in Settings → Account: what Prompteryx should call you, what best describes your work, what you want to use Prompteryx for and how you heard about us, plus which set-up steps you ticked. We use them to tailor suggestions to you and, added up across everyone, to understand who uses Prompteryx. You can change or clear them at any time.
• Workflows and prompts — the workflow definitions, prompts, schedules, documents and files you create or upload.
• Connection credentials — when you connect a third-party service, the OAuth tokens or API keys needed to act on your behalf. These are encrypted at rest, and are used only to run the connections you set up.
• Billing information — payments are processed by Stripe; we receive your plan, payment status and billing country, not your full card details.
• Showcase content — if you use Showcase: your public handle, display name and bio, the posts you publish, your comments, likes and reports, and the consent log described in section 5.3.
• Published pages and shared content — web pages, websites and apps you publish, files you share by link, workflows you share and who you share them with, and the settings you choose (such as visibility and link expiry).
• Reports — if you report content, with or without a Prompteryx account: the reason you choose, the details you give, the link (URL) of the item, your email address if you choose to give it and, if you are signed in, your account identifier. We also record a hashed (not readable) form of your IP address to stop the report tool being misused. See section 5.7.
• Personal agent — where you use it: your agent's settings, schedules and the pages you ask it to watch, and, from accounts you connect, today's and tomorrow's calendar events and recent unread emails (sender, subject and a short snippet), the briefs and proposed actions it prepares, and the delivery channels you choose (for example a browser push address).
• Agreements and confirmations — the versions of our Terms, this policy and the Acceptable Use Policy you accepted, when, your confirmation that you are 18 or over, a hashed form of your IP address and your browser's user agent; and, if you use them, your consent to desktop mode and to the App Builder Terms (and any Kimi consent you gave before 27 October 2026), and any withdrawal from a contract you send us.
• Support communications and enquiries — messages you send to contact@prompteryx.ai or through our support and Enterprise contact forms (for an Enterprise enquiry: your name, work e-mail, company, role, company size, what you are interested in and your message).

1.2 Information created by using the Service

• Run logs and recordings — when your automations run we keep execution logs, step results and (for cloud browser sessions) screenshots and session video recordings, so you can review what your automation did.
• Usage data — feature usage, credit consumption, device/browser type, IP address and similar log data, used to operate, secure and improve the Service.
• Desktop mode — if you let an AI model operate your own computer, screenshots of your screen are sent to the AI provider you choose for each step; we do not store the screenshots, only a fingerprint of each one and a record of the actions taken.
• View counts — we count how many times each published page is opened, once per visitor in any 30 minutes, using a hashed form of the visitor's IP address that we do not keep in readable form. We do not use it to identify or follow visitors.
• How you found us — when you create an account we record how you arrived: the campaign tags in the link you followed (such as utm_source or utm_campaign), the click identifier an ad platform added to it (such as Google's gclid or Meta's fbclid), the website that linked to us (its address, not what was on it), the first page you saw and when, and any referral code. We keep this with your account and copy the source and campaign to our payment provider Stripe when you buy a plan, so we can see which campaigns bring paying customers; we do not share it with advertising networks. Unless you turn on "Analytics & marketing" in the cookie banner, it is held only in the open page and sent once, with your sign-up (section 7).
• Referrals — if you invite friends with your referral link: your referral code, the people who signed up with it (shown to you only as a masked e-mail such as j•••@gmail.com), their status and the rewards. If you joined with someone's code: that code and your status. To stop people referring themselves we compare a hashed (not readable) form of the sign-up IP address and of the payment card's fingerprint from Stripe; we never see or keep card numbers.
• Request logs — when anyone, signed in or not, opens a published page, a published app or a shared file link, our hosting providers record standard request logs (such as IP address, browser user agent, time and address requested) for security and operations.

1.3 Where we get personal data from

Mostly from you. We also receive personal data from: sign-in providers you use (Google, Apple); accounts and apps you connect (only what the connection allows and your automation asks for); our payment provider Stripe (payment status, billing country, fraud signals); other users (for example when someone shares a workflow with you, mentions you in content, or reports content); and public web pages your automations visit, as you direct. Giving us your e-mail address is necessary to create an account; everything else is optional, although some features need it.

1.4 Chrome extension (Prompteryx AI Browser Agent)

The Prompteryx AI Browser Agent extension keeps your side-panel settings (AI model, thinking level, screenshot detail) and the name, e-mail address and profile photo link of your signed-in Prompteryx account in your browser, so the panel can show who is signed in. The secret that pairs it with Prompteryx Desktop is kept in the browser's session storage only (Chrome clears it when it closes) and is sent only to Prompteryx Desktop on your own computer. Only when you start a task or a recording does the extension capture screenshots and text of the page it is working on, the tab's URL and title, and the clicks, typing, scrolling and navigation the agent performs or you record. It sends them only to Prompteryx Desktop on your computer and to Prompteryx, to carry out the task; they are processed by our AI providers as described in section 3 and kept in your chat history (or as a workflow, if you save a recording) until you delete them or, for chats and runs, your plan's retention period ends (section 8). The extension does not read pages you are not running a task or recording on. We do not sell this data or use it for advertising.

2. How we use data

• To provide the Service: executing the workflows and automations you create, exactly as you direct — contract.
• To manage your account, subscription, credits and payments, and to send service messages (security alerts, renewal reminders, changes to our terms) — contract; legal obligation for consumer-law notices.
• To secure the Service: authentication, fraud and abuse prevention, rate limits, debugging — legitimate interests (keeping the Service and its users safe).
• To moderate shared content, handle reports and appeals, and enforce our Terms — legitimate interests; legal obligation under online-safety law.
• To support you when you contact us — contract or legitimate interests.
• To understand aggregate product usage and improve the Service (never by training AI models on your content) — legitimate interests.
• To measure which campaigns and links bring customers (section 1.2, "How you found us") — legitimate interests (knowing what our marketing achieves); keeping the record in your browser across visits — consent ("Analytics & marketing").
• To run the referral programme, give its rewards and prevent self-referral and card-reuse abuse — contract (the programme's terms) and legitimate interests (preventing fraud).
• To comply with the law, including tax and accounting rules, sanctions and export-control law, and lawful requests from authorities — legal obligation.
• To send product news and offers, only if you opt in — consent, which you can withdraw at any time.

We do not sell personal data, and we do not use your data for third-party advertising.

3. AI processing

When you use AI features, the content needed to fulfil your request — your prompts, relevant workflow context, and the page content or screenshots the AI Browser Agent is working with — is processed by our AI model providers (Google AI/Gemini, Anthropic, OpenAI) to generate the result. We use these providers through their paid business/API services, whose terms do not allow them to use your content to train their models; they may keep it for a limited time to detect misuse, and then delete it. We do not use your content to train AI models either. Where a model is run for us on a cloud service rather than by the company that made it, that service may also keep your prompts and the results for a limited time, only to check for abuse and keep the service safe, not to train models.

Kimi models (Moonshot AI). Kimi models are made by Moonshot AI. Most of them do not go to Moonshot AI: when you choose one — in chat, workflows, the API, the AI Browser Agent, desktop mode or the personal agent — the request goes through OpenRouter (United States) to one of a set of inference providers we have selected, trying providers in the EU first and then in the United States (Inceptron, Nebius, Together AI, Fireworks AI, Baseten, DeepInfra, Parasail, Amazon Bedrock, Modal, CoreWeave, Crusoe, Decart, Novita AI, Inference.net and Wafer). OpenRouter only uses providers that keep no copy of your content once the request is answered and do not use it to train models, and never providers in China. OpenRouter is our processor under its data processing agreement, which includes the EU Standard Contractual Clauses and the UK Addendum. Under each reply we name the model and the provider that ran it.

Some Kimi models are not yet offered by any of those providers, only by Moonshot AI PTE. LTD. in Singapore — today Kimi K3.1 Pro. If you choose one of them, everything that model works on goes to Moonshot AI: your prompts and files and, on agent features, screenshots of web pages and of your screen, e-mails, calendar entries and documents it reads, which can contain other people's personal data. We send these requests through OpenRouter to Moonshot AI's service under OpenRouter's zero-data-retention terms, so Moonshot AI keeps no copy and does not train on them. Only if OpenRouter is busy or unavailable do we send the request to Moonshot AI's own service directly; Moonshot AI's own terms then apply, and they allow it to use the content to improve its services, including training its models. Either way the data is processed in Singapore, and we have no data processing agreement or international-transfer safeguard of our own with Moonshot AI, so UK and EU data-protection rights may be hard to enforce there (section 10). As soon as providers in the EU or the United States offer such a model, we use them instead. Choosing such a model is your choice for that request: please do not use it for special-category or confidential data, or for other people's data you do not have the right to share. The chat's model picker marks Kimi K3.1 Pro as running on Moonshot AI, and every other model works as usual if you prefer not to use it. More at prompteryx.ai/legal/kimi.

Video, voice, music and sound effects. When you create one, your prompt or script and any images, clips or audio you supply are sent to the provider that runs the model you picked: Google (Veo, Gemini Omni, Lyria and Gemini voices), OpenAI (its voices), fal.ai (Wan, MiniMax, Kling and ElevenLabs voices, sound effects and music) or OpenRouter, which passes Seedance requests to ByteDance's BytePlus service (processed in Malaysia, Indonesia and/or the EU) and, as a fallback, Wan requests to Alibaba Cloud. The generated file is copied to our storage straight away. fal.ai and OpenRouter are listed in section 6; their transfer safeguards are covered in section 10.

Some models are not offered in some countries under their makers' terms (for example, Seedance is not available in the United States). To apply this, we look up the country of the IP address you connect from in an offline database (IP geolocation by DB-IP, licensed CC BY 4.0) when you open or use media models. The address is not stored for this purpose and is not sent to anyone.

If you connect your own ElevenLabs account, voice, music, sound-effect and ElevenLabs video requests you make with it go to ElevenLabs under your own ElevenLabs terms, and ElevenLabs bills you directly.

Personal agent. If you use the personal agent, it reads the calendar and recent unread email of the accounts you connect (sender, subject and a short snippet, not whole messages) to prepare your brief, and sends that content to the AI providers above (for a Kimi model, through OpenRouter as described above). It asks you before it contacts anyone, spends money or deletes anything, unless you change that, and it shows you exactly what it will send. Briefs and runs are kept for 30 days, activity and approvals for 30 days and inbox labels for 7 days. People who email you or attend your meetings have their details processed as part of your request (our legitimate interest in providing the assistant you asked for); emails the agent sends to them say they were written with AI.

Personal agent meeting notes (optional). If you paste or upload a meeting transcript for your agent, we send it once to Google's Gemini models (on Google Cloud) to pull out the work items (the meeting's title, date and participants, decisions, commitments, deadlines, open questions and a few key points), and then discard the transcript: it is not stored. Nothing is kept unless you choose to keep it, and you can untick any line first. Kept notes stay until each open item is done, then 30 days (or 30 days in all if you choose "Keep 30 days"), and never more than a year; you can delete them at any time. The agent is told not to keep health, beliefs or other sensitive details, or anything about people beyond the work items. Other people in the meeting have their names and what they said processed as part of your request (our legitimate interest in providing the assistant you asked for); please make sure they knew the meeting was recorded.

Personal agent suggestions (optional). If you switch on "Suggest routines from my accounts", once a day the agent looks at the senders and subjects of emails from the last 14 days and the titles of calendar events in the next 14 days, only in the connected accounts you tick, to suggest up to three routines (for example "tell me when a bill from this sender arrives"). It never reads email bodies or attachments for this, it only suggests, and nothing is set up until you press "Set it up". The look uses a Google Gemini model (the AI providers above) and costs a small amount of your AI credits, shown in Activity. What it found is kept until you dismiss it, and is deleted straight away if you switch suggestions off. Our basis is your consent, which you can withdraw at any time in the agent's settings.

Desktop mode. If you let an AI model operate your computer, each step sends a screenshot of your screen (or of the app you chose) to the AI provider you select — Anthropic, OpenAI or Google — which may keep it for a limited time to detect misuse. Anything visible is sent, including other people's messages, so close anything you do not want shared. We store a record of the run (the actions, the model and fingerprints of the screenshots, not the screenshots) for 90 days.

Telling people it is AI. Our chat and agents say they are AI; emails our assistants send to other people say they were written with AI; images, video and audio we generate carry machine-readable "AI-generated" information, and items you share carry an "AI-generated" label.

3.1 Memory and past chats

What is kept. The assistant can remember facts and preferences about you ("memories") — for all your chats or for one project — and an optional "About you" profile you write yourself. For each chat it also keeps a short digest (your first and latest question and a few keywords) so that past chats can be found.

How memories are made. A memory is saved when you ask the assistant to remember something, or automatically: alongside each reply, a separate small AI call (currently Anthropic's Claude Haiku, which may run on Google Cloud, with Google Gemini as a fallback) looks at your message and may save lasting facts that you stated or confirmed — never facts taken from documents, web pages, e-mails or search results, and never passwords, keys or codes, card, bank or ID numbers, criminal history, immigration status or private details about other people. It does not save health, religion, political views, ethnicity, sexual orientation, gender identity or trade-union membership unless you turn on "sensitive topics". The chat shows when a memory is saved. Memories are converted into numerical form (embeddings) by Google Gemini so that the relevant ones can be found.

How they are used. Relevant memories, your profile and — if you allow it — the titles of recent chats are added to what the AI sees when it answers you, and the AI can search your past chats (it reads their digests and, for the best matches, recent messages).

Your controls (Settings → Memory): use memories, learn automatically and reference past chats are on by default and you can turn each off; sensitive topics are off unless you turn them on; you can switch memory off for a single chat, and incognito chats neither use nor create memories. You can view, edit, pin, undo and delete memories one by one or all at once, ask the assistant to forget something, export them (JSON or text) and import memories from another assistant.

Why we may do this. Memory is part of the assistant you use (contract); learning automatically is in our and your legitimate interest in a more useful assistant, balanced by the exclusions and the off switch above; sensitive-topic memories rely on your explicit consent, which you withdraw by turning the setting off and deleting them.

3.2 Skills

Skills are instructions and reference files that tell the assistant how to do a particular job. Skills you create, upload or let the assistant draft (it asks before saving one) are stored in your account and loaded into a chat only when they are relevant. If you create a share link for a skill, a copy of that one skill can be opened and imported by anyone signed in to Prompteryx who has the link; the link and the copy are deleted after 30 days.

4. Data from connected third-party services

The Connect Hub and our integrations let you connect accounts such as Google, Microsoft, Slack, Airtable and other apps. For all data the Service obtains from a connected third-party API:

• we access it only to execute your own workflows and automations, as and when you direct — we do not read connected accounts for any other purpose;
• we never sell it;
• we never use it for advertising, and never transfer it to advertisers, data brokers or credit agencies;
• we never use it to train AI models;
• we retain it only as long as needed to run your automation and show you the results, and no longer;
• humans at Prompteryx do not read it, except with your explicit permission (for example a support request), where necessary for security or abuse investigation, or where required by law;
• you can disconnect a service at any time, which revokes our access; stored tokens are deleted when you remove the connection.

5. Showcase, published pages, apps and reports

Showcase lets you publish your creations for other users to see, like, comment on and remix. Nothing is published unless you choose to publish it, one item at a time; everything you create is private by default. An account-level "share by default" setting is off unless you turn it on, and it never publishes anything you created earlier. This section also covers web pages and websites you publish (section 5.1), reports and takedowns (section 5.7), sharing a page (section 5.8) and apps you build (section 5.9).

5.1 What other people can see

• Your public profile — your public handle, display name and bio (if you add one). Your handle is separate from your email address.
• Your published posts — the snapshot you approved on the review screen: title, description, the recipe (for example the prompt, model and settings) and a preview. AI Browser Agent tasks are published as the task prompt and an outcome summary only.
• Your comments, your posts' like, comment and remix counts, and links from remixes back to your post.
• Never shown — your email address, your connection credentials, run logs, and browser screenshots or session recordings. Anonymous posts show "Prompteryx user" instead of your handle and do not appear on your profile.

Recipients. Public posts can be seen by other Prompteryx users and, where Showcase is available without signing in, by anyone on the internet, including search engines. Unlisted posts can be seen by anyone who has the link, including anyone the link is passed on to. Public posts are licensed under CC BY 4.0 (see our Terms of Service, section 5), so others may copy them, and copies made before you unpublish are outside our control.

Published pages and websites. If you publish a page or website as unlisted or public, anyone who has or finds the link can view it without signing in. Public pages may be indexed by search engines; for unlisted pages we ask search engines not to index them. When someone shares a link to a page in a messaging app or social network, that service may fetch the page title and a preview image. If a page shows live data that one of your scheduled workflows refreshes, everyone who can view the page can see that data, so do not connect a page to data that you are not entitled to make public. Published pages carry an "AI-generated" label and a Report link. If a page you publish contains other people's personal data, you are responsible for having a lawful basis to publish it — we do not remove personal data from pages.

5.2 Automated redaction and review

When you prepare a post we build a new copy containing only the fields that type of post allows, then automatically replace things that look like secrets or personal data (for example API keys, tokens, passwords, email addresses, phone numbers, card and bank numbers, IP addresses and postal addresses) with placeholders, including your own name, email and connected-account identifiers. The original values are not stored with the post. The redacted copy is then checked by automated classifiers — TypeSafe (our "Jev" decision model) and, where used, one of the AI model providers in section 3 — which flag content that may contain personal data, secrets, or content that breaks our rules. Flagged posts may be held for review by a member of our team before they go live, and published posts may be re-checked when our rules change. These checks inform moderation; they do not produce decisions with legal or similarly significant effects on you, and you can ask for a human to review any moderation decision. Automated redaction cannot catch everything, so you must check the preview before publishing.

Before a page or website is shared we check it for passwords, keys and other secrets and for imitations of other organisations' sign-in pages, and refuse to share it if it fails; we do not redact personal data from pages. We may also use automated tools to help sort and prioritise reports (section 5.7). When we prepare images, video and audio for Explore, we remove location data (GPS), device serial numbers and personal names from the file's metadata, but keep the information that says the file was made with AI (the IPTC "digital source type" and, where the file still carries it, the provider's C2PA Content Credentials), because the law asks for AI-generated media to be identifiable.

5.3 Consent log

Each time you publish a post, change its visibility or unpublish it, we record an entry in an append-only consent log: your account identifier, the post, the action, the date and time, the version of the consent text you agreed to and of our Terms, a cryptographic hash of exactly what was published, and a hashed (not readable) form of your IP address together with your browser's user agent. We keep this to evidence that you gave, changed or withdrew consent, and to deal with disputes, abuse and legal claims. It is kept while your account exists and for 6 years after it is deleted, and is not shown to other users.

5.4 Lawful bases

• Consent — publishing a post and making it visible to others. You can withdraw consent at any time by unpublishing the post or making everything private; withdrawal does not affect the lawfulness of processing before you withdrew it.
• Legitimate interests — redaction, automated review, moderation, rate limits, ranking, the consent log, view counts, the sharing record, and keeping Showcase, published pages and apps secure and free of illegal content and abuse; handling reports from anyone (with or without an account), sending statements of reasons and handling appeals.
• Legal obligation — complying with the UK Online Safety Act 2023 (including our duties to run reporting and complaints procedures and to take down illegal content), responding to takedown notices and lawful requests from authorities, and reporting crimes where the law requires it. Where we handle information about suspected criminal offences, we do so under the conditions in the Data Protection Act 2018 (for example, preventing or detecting unlawful acts).
• Performance of a contract — providing likes, comments, remixes and your public profile as features of the Service you use.

5.5 Retention

• Published snapshots, previews and thumbnails — kept while the post is published; deleted within minutes when you unpublish it, make everything private, or delete your account.
• Comments — kept while the post they are on is published. Comments on a post that is unpublished are hidden immediately and deleted after 30 days. If you delete your account, your comments on other people's posts remain but are anonymised as "[deleted user]".
• Likes — kept until you remove the like, the post is unpublished or your account is deleted.
• Reports, takedown decisions, statements of reasons and appeals — including reports from people without an account and the hashed IP address recorded with them — kept for 2 years, unless we need them for a legal claim, the law requires us to keep them longer, or law enforcement asks us to preserve them. Moderation actions are also recorded in our security audit log, which stores hashed identifiers.
• Content we take down — removed from view immediately; a page we take down stays stored but unavailable, so it can be restored if an appeal succeeds, until you delete it or your account. Child sexual abuse material is handled only as needed to report it and in line with law-enforcement guidance.
• Published pages, websites and apps — kept until you unpublish or delete them, or delete your account; view counts are deleted with the page.
• Remixes made by other users are copies in their accounts and are not deleted when you unpublish; they stop linking to you.
• Consent log — see section 5.3.

5.6 Your choices and rights

You can publish, change visibility, unpublish, make everything private, edit your public profile and delete comments at any time in Settings → Sharing & privacy, or by emailing contact@prompteryx.ai. Your rights of access, rectification, erasure, restriction, objection, portability and withdrawal of consent (section 11) apply to Showcase data in the same way. Anyone — including people who do not have an account — can ask us to remove a post, comment or published page that contains their personal data by using the Report link or emailing us.

5.7 Reports, takedowns, statements of reasons and appeals

Anyone can report content using the Report link on any unlisted or public item, at prompteryx.ai/report, or by emailing contact@prompteryx.ai. We collect the information listed in section 1.1 ("Reports") and use it to assess and act on the report, to contact you about it if you gave an email address, to stop the report tool being misused (for example rate limits using the hashed IP address), and to keep a record of what we decided and why. Reports often contain information about other people — the person who shared the content, or people shown in it; we use it only to handle the report and any appeal.

If you give an email address when you report something, we send one acknowledgement to it. The message does not repeat what you wrote, and we send at most a few acknowledgements to any one address per day, so the form cannot be used to send email to other people. Our moderation team and the service providers who help us (section 6) can see reports. We do not tell the person whose content was reported who reported it, unless that is strictly necessary (for example, an intellectual-property complaint where the rights holder has to be identified).

When we act on content we send its owner a statement of reasons and keep a record of the decision, the statement and any appeal; if you appeal, we keep your message, the outcome and the reasons with the decision record. If you delete your account, reports and appeals you filed stay as a record of what was reported, but we remove your account identifier, email address, name and the hashed IP address from them. We may share a report, and information about the account concerned, with law enforcement or other authorities where the law requires or allows it — for example, reporting suspected child sexual abuse material to the National Crime Agency, the Internet Watch Foundation or the US National Center for Missing & Exploited Children, or telling the police about a threat to someone's life or safety.

5.8 Sharing a page

When you share a web page or website, we record that you accepted the sharing statement shown at that moment: its version, a fingerprint of its exact text, the date, your browser's user agent and a hashed version of your IP address. We keep this record with the page while it exists, and in our security audit log. Lawful basis: legitimate interests (being able to show who shared what, and on what terms).

5.9 Apps you build

If you build an app with the App builder, the people who use your app ("app users") sign in and store data in your app, not in a Prompteryx account. You are the controller of your app users' personal data and we process it on your behalf as your processor, under the data processing agreement in our App Builder Terms: we store it, authenticate your app users (we keep only a hash of their password), run your app's functions and AI features, and serve your published app. Your app should have its own privacy notice; app users with questions about their data should contact the app's owner. We process limited data about app users as a controller only for security, abuse prevention and handling reports about an app. When you delete an app or your account, its app users' data is deleted.

6. Subprocessors and service providers

We share data only with the providers we need to run the Service, under contracts that limit their use of it, except where an entry below says otherwise:

We may also disclose data where required by law, to protect the rights and safety of users or the public (for example the reports to authorities described in section 5.7), or as part of a business transfer (in which case this policy continues to apply to it). For the personal agent, browser push notifications are delivered through your browser's push service (for example Google, Mozilla or Apple) in encrypted form.

7. Cookies

We use strictly necessary cookies to keep you signed in and protect the platform (authentication, CSRF/security). Functional and "Analytics & marketing" cookies are off by default and only set if you enable them in the cookie banner, where you can also change your choice later. If you are signed in, we also save your choice to your account. We use no third-party analytics, advertising cookies or ad pixels of any kind.

Analytics & marketing does one thing today: it lets your browser keep a record of how you first found us (the campaign tags, ad click identifier, referring website and referral code described in section 1.2) for up to 90 days, so that if you sign up on a later visit we still know which campaign brought you. It is kept by us only, removed when you sign up or turn the setting off, and never sent to an ad network. With it off, we keep nothing on your device: the details of your current visit stay in the open page only and are sent once if you create an account during that visit. Referral links carry the code in the link itself, so they need no cookie.

Published pages, websites and apps are their publisher's own pages, not our app, and we do not set cookies on published pages. If a page embeds content from another service (for example a YouTube video or a Google map), that service may set its own cookies, so the embed loads only after the visitor clicks "Load". Apps people build may set their own storage; their owners are responsible for it.

8. Data retention and deletion

• Account data and workflows — kept until you delete them or your account; you can delete individual workflows at any time in the app.
• Chats, run history and downloads — chats (counted from the last message), agent and workflow runs with their logs and step results, generated images and captured downloads are kept for your plan's retention period and then deleted automatically: Free 7 days, Developer 30 days, Starter 90 days, Builder 90 days, Pro 365 days, Enterprise 730 days (Enterprise: or as agreed in your contract). Chats that an active schedule still posts into, and files you have published, are kept while that is so. You can delete individual chats, runs and files sooner in the app.
• Cloud browser session recordings — deleted automatically 14 days after the session (sooner if you delete them), whatever your plan.
• Connection credentials — deleted when you remove the connection.
• Personal agent — runs and briefs 30 days, activity and approvals 30 days, inbox labels 7 days; meeting notes until each open item is done and then 30 days (at most a year; transcripts are not kept); settings, schedules and watched pages until you delete them.
• Desktop mode — run records 90 days; screenshots are not stored.
• Memory and "About you" — kept until you delete them (Settings → Memory) or your account. A memory is not deleted when the chat it came from is deleted under your plan's retention period; the digest of a chat is deleted with the chat.
• Skills — kept until you delete them or your account; skill share links and their copies are deleted after 30 days.
• How you found us and referrals — the sign-up source is kept with your account until you delete it. A referral record is kept while both accounts exist and is deleted when either account is deleted; the record in your browser (only with "Analytics & marketing" on) expires after 90 days.
• Enquiries — Enterprise enquiries sent through our contact form are kept for 2 years, so we can follow up and keep a record of what was discussed.
• Agreement records — the record of the Terms versions you accepted is kept while your account exists; a withdrawal from a contract is kept for 6 years as evidence.
• Showcase posts, comments, likes, reports and the consent log — see section 5.5.
• Account deletion — you can permanently delete your account (Settings → Account → Delete account on the web, or Settings → "Delete my account…" in the mobile app) or ask us at contact@prompteryx.ai; step-by-step instructions and the full list of what we keep are on our Delete your account page. Deletion removes your personal data, workflows, credentials and recordings within 30 days, except minimal records we must keep for legal, billing or security reasons (including the Showcase consent log, section 5.3). It also unpublishes and deletes all your Showcase posts within minutes, deletes your public profile and likes, and anonymises your comments on other people's posts as "[deleted user]".
• Security and audit logs — kept for 7 years, to protect the Service and meet our legal and compliance obligations. Other logs expire on rolling schedules and are then removed automatically.
• Backups — deleted data can remain in our backups for up to 14 days before those backups expire.

9. Security

• Encryption in transit (TLS) for all traffic, and encryption at rest for stored data.
• OAuth tokens and API keys are additionally encrypted at rest and never exposed to other users or in client-side code.
• Cloud browser sessions are isolated per user session and destroyed when the session ends.
• Access to production systems is restricted, logged and reviewed.

No internet service can guarantee absolute security; if we learn of a breach affecting your personal data we will notify you and the relevant authorities as the law requires.

10. International transfers

We operate from the United Kingdom and use infrastructure in the UK, EU and United States. Where personal data is transferred out of the UK/EEA we rely on appropriate safeguards — adequacy regulations, the EU Standard Contractual Clauses and the UK Addendum/IDTA — with the providers listed in section 6.

For fal.ai (video and audio generation, section 3) we are putting in place its Data Processing Addendum, which incorporates the EU Standard Contractual Clauses and the UK International Data Transfer Addendum. OpenRouter's data processing agreement (part of its terms) incorporates the EU Standard Contractual Clauses (controller to processor) and the UK Addendum; that covers Kimi requests and the inference providers OpenRouter uses for them. For BytePlus behind OpenRouter for Seedance we are still confirming the transfer safeguards; until then, your content is sent to it only for the Seedance requests you make.

Moonshot AI (Singapore), for the Kimi models only it serves (section 3), is a transfer without any of our own safeguards: there is no adequacy decision for Singapore, and no data processing agreement, Standard Contractual Clauses or IDTA between us and Moonshot AI. Through OpenRouter it acts under OpenRouter's zero-data-retention terms; on the last-resort direct route it acts under its own terms. Your content goes there only for requests you make with such a model.

11. Your rights and our lawful bases

We process personal data under UK GDPR/EU GDPR on these bases: performance of a contract (providing the Service you signed up for), legitimate interests (securing and improving the Service, preventing abuse, moderating Showcase), consent (optional cookies, including "Analytics & marketing"; marketing emails; publishing to Showcase — which you can withdraw at any time, without affecting processing before you withdrew), and legal obligation (tax, accounting, lawful requests, takedown notices). Section 5.4 sets out the bases for Showcase in more detail, and section 3.1 those for memory (including explicit consent for sensitive-topic memories). You can object to automatic memory learning at any time by turning it off.

You have the right to:

• access the personal data we hold about you, and receive a copy (you can export your data from the app);
• correct inaccurate data;
• delete your data (see section 8);
• restrict or object to processing, including objecting to legitimate-interest processing;
• data portability;
• withdraw consent where processing is based on consent; and
• complain to us — email contact@prompteryx.ai with the subject "Privacy complaint"; we acknowledge complaints within 30 days and tell you the outcome; and
• complain to a supervisory authority — in the UK, the Information Commissioner's Office (ico.org.uk); in the EU, the authority where you live or work.

To exercise any of these rights, email contact@prompteryx.ai. We respond within one month (we may extend this by up to two further months for complex requests, and will tell you if we do). We may need to confirm your identity first.

11.1 Automated decisions

We do not make decisions about you based solely on automated processing that have legal or similarly significant effects, except these safeguards against abuse: automated checks at sign-up (for example disposable e-mail addresses or unusual sign-up patterns) can stop an account being created, and our payment provider’s fraud screening can decline a payment. If that happens to you, contact us: a person will review the decision, you can explain your point of view, and you can contest it. Content moderation tools can hide items temporarily, but a person makes any final decision (Terms of Service section 5.9).

12. Children

The Service is not intended for anyone under 18, and we do not knowingly collect personal data from children. You confirm you are 18 or over when you create an account. If you believe a child has created an account, contact us and we will delete it. Published pages can be opened by anyone without an account; if you are worried about something you have seen, report it using the Report link on the page.

13. Changes to this policy

We will post any changes on this page and update the "Last updated" date; for material changes we will also notify you (for example by email or an in-product notice) before they take effect.

14. Contact