Privacy policy
DRAFT — NOT PUBLISHEDThis text was drafted by the gp-v2 worker under proj010 · task-p8-privacy-tos-dsr, with the data-protection questions routed through the security_consult seat (consult_mttpl0xrq1iaw1, consult_mttppev6oi0ere, 2026-09-09) before it was written. It is not law and it has not been reviewed by a lawyer. /privacy renders its honest-empty until the founder approves this file — residual task-p8-legal-approve.Every 〔…〕 is an open decision only the founder can close. They are listed again at the end so none is closed by being overlooked. When this file is approved, its sha256 goes into node/src/dsr/policies.ts as POLICIES.privacy.textSha256, and that is what makes consent to it recordable at all — until then the node refuses to create an account, on purpose.Version 2026-09-09 · draftGreenplateAi is a kitchen companion: it helps you plan meals, cook them, keep a pantry and a shopping list, and log what you ate. This page says what we hold, why we are allowed to hold it, who else touches it, how long it stays, and what you can make us do about it.We have written it to be read, not to be survived. Where something is uncomfortable — that diet data implies health data, that erasure does not delete absolutely everything, that an AI feature sends your words to another company — it says so in the same plain register as the rest.
1. Who we are
GreenplateAi is operated by 〔legal entity name, registered address, company number〕.For the purposes of the UK GDPR and the EU GDPR, that entity is the data controller for everything described here.Contact for anything on this page: 〔privacy contact address — proposed privacy@greenplate.ai, which needs an actual mailbox routed to it before this file is published; a policy that names an address nobody reads is worse than one that names none〕.〔Whether an EU/UK representative under Art.27 is required, and a Data Protection Officer under Art.37 — both depend on where the entity is established and how many users are in the EU/UK. To be settled before launch, not after.〕
2. What we hold, and the lawful basis for each
Not one blanket sentence. Each activity has its own basis, because they are genuinely different and because you have different rights depending on which one applies.What we process · Examples · Lawful basis · Can you refuse it and still use GreenplateAi?Your account · a one-way fingerprint of your email address, your sign-in identity at Clerk, when you joined · Contract (Art.6(1)(b)) · No — there is no account without itYour kitchen · meal plans, pantry, shopping list, cooking log, favourites, saved recipes, taste profile · Contract · No — this is the productBilling · your Stripe customer and subscription reference, your plan and its status · Contract, and legal obligation (Art.6(1)(c)) for the tax record · No, if you subscribeSecurity and abuse logs · one line per action: when, which tool, whether it succeeded, how long it took · Legitimate interests (Art.6(1)(f)) — keeping the service safe and available · No — but see §6 for what it does and does not containUsage counters · how many requests and how much AI capacity you have used today · Contract and legitimate interests · No — they are how the free tier and the abuse ceiling workAI features · the meal/diet/allergen information used to answer *your* request, sent to an AI model provider · Consent (Art.6(1)(a)), at the heightened standard in §3 · Yes. Refuse it and everything else keeps workingProduct analytics · which screens and features are used, via PostHog · Consent (Art.6(1)(a)) · Yes. Refuse it and nothing else changesThe two things that follow from this table.Consent means *consent*: for the AI features and for analytics you may say no at sign-up, and you may change your mind later, and the rest of the service — your account, your kitchen, your subscription — is unaffected either way. If you withdraw consent to the AI features, the AI features stop; your subscription does not silently cancel and your data is not deleted.And the reverse: your account, your kitchen and your billing are not consent-based. We do not show you a consent box for them, because you cannot withdraw them and keep the service, and a checkbox that pretends otherwise would be dishonest. If you want that processing to stop, the way to stop it is to close your account — §7.
3. Meal data is health-adjacent, and we treat it that way
What you eat reveals things about you that you may not have thought of as medical: allergies and intolerances, a diagnosed condition you are eating around, a pregnancy, a disorder, a religious observance, an ethical commitment.We do not ask you to declare any of that, and GreenplateAi has no "health record". But a pantry, a meal plan and an allergen list can imply it, so we handle that information at the standard the GDPR reserves for special-category data (Art.9) rather than the ordinary one:• Sending it to an AI model provider needs your explicit, separate, opt-in consent. It is never bundled into accepting the terms, never pre-ticked, and refusing it is a normal answer we record as such.• We do not use it to infer or store a health status, and we do not sell or share it for advertising or profiling. Ever, under any consent.• It is not used to train anybody's AI model — ours or a provider's. See §5.• 〔If any feature is ever added that infers a condition from meal data, it needs its own consent and its own row in the table in §2, and this paragraph must be rewritten rather than stretched.〕
4. Your data does not leave your own household
GreenplateAi is built so that one household's data is unreachable from another's — not by policy, by construction. Every record carries the household it belongs to, the database refuses a query that does not name a household, and the connection the service runs on has no privilege to look past that. A request that has not established who it is sees nothing at all.Our own staff are subject to the same wall. An operator cannot browse your kitchen, your account or your consent record; reaching a specific household requires a specific, logged action, and the log records the operator who did it rather than pretending it was you.Your data also stays inside GreenplateAi's own system. It is not pooled with, or reachable from, any other business run by the same people or on the same platform.
5. Who else touches your data
These are the only companies that process your data on our behalf. Each is under a data processing agreement 〔to be executed and dated before launch — one line per processor〕.Processor · What it does · What it can see · WhereMicrosoft Azure · hosts the service and the database · everything stored, as our infrastructure provider · 〔region — currently East US; if EU/UK users are served from there, §9 applies〕Cloudflare · serves the website and stores recipe images · page requests; your own uploaded images if you add any · global edgeClerk · sign-in and accounts · your email address, sign-in method, session activity · 〔region〕Stripe · payments and subscriptions · your name, payment details and billing address — we never receive or store your card details · 〔region〕Anthropic 〔and/or the model provider actually wired — to confirm〕 · answers your AI requests · the text of your AI request and the recipe context it needs — only if you consented · 〔region〕PostHog · product analytics · which features you use, on a pseudonymous identifier — only if you consented · 〔region / self-hosted〕Resend 〔if used〕 · sends service email · your email address and the message · 〔region〕About the AI provider, specifically. When you use an AI feature, your request and the recipe context needed to answer it are sent to the provider named above. We contract with them so that your content is not used to train their models, and we do not use it to train any model of our own. 〔Confirm the exact no-training and retention terms in the provider's DPA and state their log-retention period here — a policy that says "not used for training" without naming the term it relies on is a promise with nothing behind it.〕About analytics, specifically. PostHog is configured so that automatic capture does not collect the contents of form fields. Your meal plans, pantry contents, allergen notes and messages to the assistant are not analytics events. 〔This is a configuration, so it needs a check in the deploy — an unconfigured autocapture will scoop exactly the fields this sentence promises it does not.〕
6. How long we keep things
Concrete periods, because "as long as necessary" is not an answer.What · How long · WhyYour account and kitchen data · until you delete your account, then per §7 · it is your working dataConsent records · for the life of the account, plus 〔6 years〕 after · we have to be able to *prove* what you agreed to and when (Art.7(1))Billing and tax records · 7 years 〔confirm per jurisdiction〕 from the transaction · tax and accounting law requires it, and that requirement outlives an erasure request (Art.17(3)(b))Security and abuse logs · 〔12 months〕 while your account is open; 90 days in pseudonymised form after erasure, then deleted · fraud, abuse and incident investigation (Art.6(1)(f); Art.17(3)(b) and (e))Usage counters · 〔13 months〕 · capacity planning and abuse detectionErasure tombstone (see §7) · 〔7 years〕, alongside the billing record · so a later payment webhook or sign-up cannot silently rebuild a deleted accountBackups · 7 days (point-in-time recovery) · a deletion reaches backups as they age out, not instantly — see §7
7. Your rights, and exactly what we do when you use them
You can see your data, get a copy of it, correct it, delete it, object to processing based on our legitimate interests, and withdraw any consent you have given. Withdrawing consent is as easy as giving it, and does not affect anything we did while it was valid.We answer without undue delay and within one month (Art.12(3)). If a request is genuinely complex we may extend by two months, and we will tell you why inside the first month.Using any of these rights costs nothing and does not degrade your service.Getting a copy of your data (Art.15 and Art.20)Ask in the app, or through the connector, or by email. You get a machine-readable file containing your account record, your consents, your rights requests, your entitlements, your Stripe references, your subscription state, your usage counters and your activity history.Two deliberate limits, stated rather than hidden:• It is delivered to you out of band — by a link or an email to your verified address, never through the AI assistant. Returning your personal data inside an assistant's reply would put it through a model provider's context and logs to answer a question you asked about yourself, which is not a trade we will make on your behalf.• One field of the activity log is withheld: a cryptographic fingerprint of each request's arguments. It is not readable information about you — it is an integrity check — and disclosing it would mainly help someone who had stolen your session work out how to tamper with the log. Everything else in the log is included. 〔Art.15(4).〕The export file itself says what it contains and what it leaves out, so you can check this paragraph against the thing you actually received.Deleting your account (Art.17)You ask, and three things happen in this order:1. Your access ends immediately. Not in thirty days — at the moment you ask.2. A 30-day grace period runs, during which you can 〔contact us to〕 undo it and during which we cancel your subscription cleanly. Nothing is deleted yet.3. Then the data is deleted: your kitchen, entitlements, Stripe references, subscription record, usage counters and consents.What survives, and why. We would rather tell you this than have you discover it.• Billing and tax records, for the period in §6, because tax law requires it. They are removed from your account and from any future export.• Security log entries, pseudonymised — the record that an action happened, with the identifier that pointed at you replaced by one that does not — for 90 days, then deleted.• A tombstone: the one-way fingerprint of your email address and your Stripe customer reference. It exists so a payment webhook arriving after you left, or a later sign-up with the same address, cannot quietly rebuild the account you deleted. Without it, "deleted" would not stay deleted.• Backups hold your data until they age out (7 days), because a backup you can edit is not a backup. Restored data is re-deleted.You are welcome back afterwards. A new sign-up with the same email address gets a new, empty account — the old one is never reattached.ComplainingIf we get this wrong, tell us first — 〔contact address〕 — and if you are not satisfied you can complain to your data protection authority. In the UK that is the Information Commissioner's Office (ico.org.uk); in the EU it is your national supervisory authority. Complaining to them does not require asking us first.
8. Security
The service runs on a private database with no public endpoint, reached over an encrypted connection by a role with no privilege to read across households. Your email address is stored as a one-way fingerprint rather than as an address. Secrets live in a managed vault, never in code or configuration. Every action is logged. Payment card details never reach us.〔Add: whether an independent security assessment has been done, and the address for reporting a vulnerability. Both are cheap and both are asked for by app stores and connector directories.〕
9. International transfers
〔The stack is hosted in 〔region〕 and several processors in §5 operate outside the UK/EEA. If GreenplateAi serves UK or EU users, this section must name the transfer mechanism relied on for each — the UK IDTA/Addendum, or the EU Standard Contractual Clauses, or an adequacy decision — and say that a copy is available on request. This is the section most often left out, and it is not optional.〕
10. Children
GreenplateAi is not for children. You must be 〔16 / 13 — the digital-consent age varies by country and the choice sets the wording〕 or older to have an account. We do not knowingly collect data from children; if you believe a child has an account, tell us and we will delete it.
11. Cookies and similar technologies
The website uses what it needs to work (sign-in and security), which needs no consent. Anything else — analytics — is off until you turn it on. 〔Confirm the actual cookie set and list it here, or link a cookie page. A policy that describes a consent banner the site does not have is a finding.〕
12. Changes to this policy
Every version has a date, and we keep the old ones.When we change something material — a new processor, a new purpose, a new use of your meal data — we ask you again before we rely on the change, and any consent you had given for the affected purpose stops applying until you answer. The feature pauses; your account, your data and your subscription do not.When we change something immaterial — a clarification, a typo, a reworded sentence — we bump the version and tell you, and nothing pauses.We classify each change one way or the other when we publish it, rather than deciding after someone objects.
Open decisions 〔founder〕
Nothing below is a drafting gap — each is a fact or a choice that only the founder can supply, and each is load-bearing.1. Legal entity name, registered address, company number (§1).2. A privacy contact address with a mailbox that is actually read (§1).3. Whether an Art.27 representative and/or an Art.37 DPO are required (§1).4. Hosting and processor regions, and the transfer mechanism for each (§5, §9). This is the section most likely to be wrong at launch.5. The AI provider actually wired, and its no-training and log-retention terms as written in its DPA (§5).6. Executed DPAs for every processor in §5, dated.7. The confirmed tax-retention period for the entity's jurisdiction (§6).8. Minimum age (§10) and the cookie set (§11).9. A vulnerability-reporting address (§8).10. Legal review of the whole file. This draft is a starting point for that review, not a substitute for it.
This page is the draft text you accept when you create an account. The exact text, byte for byte: /legal/privacy-draft.txt