Replyz Privacy Notice
Version 2026-11-10-en
This page never changes: a new version gets a new address. Same version: Terms of Service, version 2026-11-10-en
SHA-256 of this version's texts (Terms and Privacy Notice together): 39ee65c125c814026206664791008ec558031eb90a4c4f20f074c51705a326ab
# Replyz — Privacy Notice
> Version: **2026-11-10-en** · In force from 10 November 2026 · Language: English (a Portuguese version exists)
This notice explains how Wonder Flamingo, Unipessoal Lda ("Replyz", "we") handles personal data when you use Replyz, and
when you send an email to a Replyz address. It follows the General Data Protection Regulation (GDPR) and Portuguese law
(Lei 58/2019).
## 1. Who we are and how to reach us
- Wonder Flamingo, Unipessoal Lda, Rua Casal do Olival 10, 2745-739 Massama, Portugal; single registration and tax number 514097400.
- Privacy questions and requests: privacy@replyz.io.
- Abuse reports: abuse@replyz.io.
- We have not appointed a Data Protection Officer: the law does not require one for our processing (GDPR art. 37).
## 2. Whose data this is about
- **Account owners**: the person who claims an account (the human behind an agent).
- **Agents**: an agent is a program, but what it sends us (for example the name it gives itself) may say something about its
owner.
- **Senders**: anyone who sends an email to a Replyz address (for example a service sending a verification code, or a
person writing to an agent). Section 9 is written for you.
## 3. Our role
| Data | Our role |
| --- | --- |
| Account data, security and abuse data, technical records, email from your agent to you (section 4, items A–D and F) | **Controller** |
| Content of the inboxes of a **claimed** account (emails, attachments, extraction results) | **Controller**: we decide how mail is received, checked, stored and deleted. What your agent then does with your mail is your own business |
| Content of the inboxes of a **sandbox** account (no human has claimed it yet) | **Controller**, on the basis of legitimate interest (section 5) |
## 4. What we collect
**A. Sign-up and account**
- The name the agent gives itself (`agent_name`), the account ID, creation date, sandbox end date and account status.
- A keyed hash (HMAC) of each API key. We never store the key itself and cannot recover it.
- A keyed hash of the IP address (or, for IPv6, of its /64 network) used to sign up. **We never store or log the IP address
itself.** The hash is used to count sign-ups per network.
**B. Claim and manage**
- The email address you claim with, the version of the terms you accepted and the time of the claim.
- A hash of the one-time code shown on the claim or manage page, and its expiry.
- The email you send to prove the claim, which is stored as a hidden system message in the agent's inbox.
- If you join the optional "Founding Builder" list on the claim page: the time, and, if you answer, how much you would
pay per month (a number of euros), stored with your account.
- **If you use the owner dashboard** (replyz.io/manage, after you prove your address with one email; the dashboard may be
available in English only): a session, kept by a strictly necessary cookie that holds a random identifier (we store only a
hash of it). It ends after 30 minutes without use, or 8 hours after you signed in to the dashboard, or when you end it. We also
keep an **owner-view log**, for security, so that a stolen session can be seen. It records when a session starts and ends
(sign-in, sign-out including "end all sessions", or timeout), with a keyed hash of your network (as for sign-ups; we never
store the IP address itself), and each time you open a message, view a message's source, open the copy of an email from
your agent, or download an attachment. Each line holds the account, the time, the kind of event, a hash of the session and
the ID of the message, attachment or email: never the content.
- Your dashboard settings: whether you want the weekly summary, risk alerts (and from which risk level) and the 80% warning,
and the pause switches you set (receiving, keys, inboxes, emails from your agent to you, creating new inboxes), with the time each was set.
- Sender lists you set in the dashboard (block or allow): the addresses or domains you enter, at most 100 per account. Mail
from a blocked sender, or, if you set an allow list, from any sender not on it, is rejected with "mailbox unavailable" and
is not stored.
- The **activity timeline** the dashboard shows is its own record, kept 30 days (at most 20,000 rows per account): what your
agent did (waits, whether a code or link was found, never its value, inboxes created or deleted, limits reached), your own
changes in the dashboard, and, for mail that was rejected, only the sender's **domain** and the reason. It holds internal
IDs and fixed codes only.
**C. Security and abuse**
- Counters (for example sign-ups per IP hash per day, messages received per account per month, rejected deliveries by
reason).
- Abuse records about an account (for example a burst of sign-ups or a flood of one-time-code emails), with the account ID
and a short description.
- A keyed hash of every inbox address that ever existed, kept permanently so that an address is never given to anyone else.
**D. Technical records**
- Our operational logs contain internal IDs, fixed status codes and timings only: no message content and no email
addresses. They are stored in the EU (Cloudflare D1, EU jurisdiction) and kept for 14 days. Cloudflare's Workers Logs is
turned off.
- **Cloudflare's Email Routing activity log.** Cloudflare, which receives the mail for us, keeps a log of every email sent to
a Replyz address, accepted or rejected, with the **sender, recipient and subject**. We have found no setting to turn it off.
Cloudflare does not document how long it keeps it or where.
**E. Inbox content**
- Each email received: headers (sender, recipients, subject, dates, message IDs, authentication results), text and HTML,
attachments, and the original message as received (raw MIME).
- What we derive from it: a cleaned text, a short preview, a safety score and signals (section 6), whether the sender could
be authenticated, Cloudflare's spam score, a "looks like it carries a code" flag, and the codes and links found when the
agent asks.
- When your agent asks for an attachment as text, the text we get by converting it (section 6), stored with that attachment.
- Threads, and the time an email was first returned to the agent (the "read" mark).
- A list of senders that passed authentication, per inbox ("contacts").
- Events (for example "message received"), with the sender and subject, so the agent can follow new mail.
**F. Email from your agent to you**
- The subject and text your agent writes to you. They wait in an EU queue until the email is sent. From the moment your agent
queues it, **we keep a copy for 12 months** (including emails that end up not being sent), so that you can read it in your
dashboard ("Emails to you") with its status. The email goes only to the address you
claimed the account with. There is no delete button for a single email; you can delete the whole account, or ask us at
privacy@replyz.io to erase one.
- A count of these emails per account per day (at most 10).
- **Emails from Replyz itself** about your account: the weekly summary, risk alerts and the 80% warning. They contain counts
only, never a subject, sender, code or message text. They are service messages, not marketing, and you can turn each of
them off in the dashboard (the legal notices we must send you cannot be turned off). We keep a record of each one sent
(kind, time, numbers and internal IDs, a hash of your address and the language) for 12 months.
- Delivery events from Cloudflare (bounce, spam complaint, rejection, failure), with the recipient and subject. They arrive
in a queue of ours in the EU and are kept for up to 1 day.
- If an email to you bounces, is rejected or is marked as spam: your address on our suppression list (so we stop emailing
it), and a pause on your account's emails to you, with the reason.
- **Cloudflare's Email Sending records.** Cloudflare, which sends these emails for us, keeps an activity log of the emails
sent (about **30 days**) and sending metrics (**31 days**). Cloudflare does not say where they are stored. We have turned
off Cloudflare's message preview. Cloudflare does not document exactly which fields it keeps.
The only cookie we set is the dashboard session cookie described in item B, and only after you sign in to the dashboard. It is
strictly necessary to give you the service you asked for, so it needs no consent. We use no trackers, analytics or advertising
tools on replyz.io, docs.replyz.io or our other domains. On docs.replyz.io, if you choose a light or dark theme, your browser
remembers that choice on your device; it is never sent to us. This needs no consent: you asked for it.
## 5. Why we use it and on what legal basis
| Purpose | Data | Legal basis (GDPR art. 6(1)) |
| --- | --- | --- |
| Providing the service to a claimed account: receiving, storing and showing mail, wait, extraction | E, A | Performance of the contract (b) |
| Providing the service to a sandbox account before anyone has claimed it | A, E | Legitimate interest (f): providing the inbox the agent asked for, so its owner can try the service and claim it |
| Proving and recording the claim, and the terms accepted | B | Performance of the contract (b); keeping proof of acceptance: legitimate interest (f) |
| Revoking keys and deleting the account on request through the manage page | B, A | Performance of the contract (b) |
| Learning what users would pay, from the optional question on the claim page | B (your answer) | Legitimate interest (f): deciding whether and how to offer a paid plan. Answering is optional |
| Sending you the emails your agent writes to you, and showing you a copy for 12 months | F | Performance of the contract (b) |
| Showing the owner of a claimed account the mail the account receives and the emails it sends, in the dashboard, with the settings, lists and timeline | B, E, F | Performance of the contract (b) for the owner; legitimate interest (f) for senders: you wrote to an address that belongs to the owner, whose agent already passes the mail on (section 9) |
| The weekly summary, risk alerts and 80% warning to the owner (service messages, counts only) | B, F | Performance of the contract (b) |
| Keeping the owner-view log and the dashboard session secure | B | Legitimate interest (f): seeing a stolen session; security of processing (GDPR art. 32) |
| Converting an attachment to text when the agent asks | E | Performance of the contract (b) for claimed accounts; legitimate interest (f) for sandbox accounts and for senders |
| Stopping emails to an address that bounced or complained (suppression list, pause) | F | Legitimate interest (f): not emailing people who do not want or cannot receive the mail, and protecting the sending domain |
| Preventing abuse and keeping the service secure: proof of work, limits, IP hash counters, abuse records, reject counters, flood flag, suspensions | A, C, E (counts and flags only) | Legitimate interest (f): protecting the service, its users and the people who receive email from being harmed by spam, fraud and mass account creation |
| Never reissuing an address | C (address hashes) | Legitimate interest (f): stopping mail meant for a past owner reaching a new one |
| Safety checks on inbound mail (section 6) | E | Performance of the contract (b) for claimed accounts; legitimate interest (f) for sandbox accounts and for senders |
| Answering requests and complaints | Contact data, the request | Legal obligation (c) where the law requires an answer; otherwise legitimate interest (f) |
You can object to processing based on legitimate interest (section 10).
## 6. Automated processing of email content
Every email that arrives is processed automatically, without anyone reading it:
- hidden text and invisible characters are removed from the text given to the agent;
- the message is scored for signs of instructions aimed at AI agents (prompt injection): `low`, `medium` or `high`, with the
signals found (for example "hidden instructions");
- the sender's authentication result is read from Cloudflare's mail servers' checks;
- messages that look like they carry a one-time code are marked, and codes and links are found when the agent asks;
- messages are counted per account, and an account that suddenly receives many code emails from many sender domains is
flagged; on a sandbox account this lowers the number of extractions per hour for 24 hours, and we are alerted;
- attachments are not scanned for viruses or other malware; attachment types that are often dangerous are marked, and
attachments are always offered as downloads;
- **only when your agent asks for an attachment as text**, the attachment (PDF, Word, spreadsheet, CSV, HTML or XML) is sent
to Cloudflare's Workers AI service to be converted into text. Cloudflare may use AI models for this, for example for images
inside a document. Only the converted text is kept, stored with the attachment and deleted with it (section 11). Reading
text from images (OCR) is not turned on. Cloudflare states that it does not use customer content to train AI models and
that it keeps inputs and outputs only as long as needed to provide the service, except where the law requires otherwise.
Cloudflare does not say where this processing runs (section 7).
The checks above are rules written by us (heuristics). They do not use AI models; the attachment conversion is the one step
where an AI model of our provider may be involved, and it only changes the format of the attachment, it decides nothing. They do not make decisions with legal or similarly
significant effects on anyone (GDPR art. 22). Links in emails are never opened and images are never loaded.
We do not use the content of your inboxes to train AI models, and we will not start doing so without your express consent.
Nobody at Replyz reads inbox content (the owner of a claimed account reads it in the dashboard, see section 4.B), with one exception: the Director may look at the content of an account while
investigating an abuse report about it. Such reads are done only through our own support tool, which logs each one (who,
when, which account, which report). The Director also has technical access to the underlying storage, which the tool
cannot record; our internal rule is that content is never read that way.
## 7. Where your data is
- **At rest, all customer data is stored in the European Union**: in Cloudflare's database (D1), object storage (R2), queues
and Durable Objects, all created in Cloudflare's EU jurisdiction.
- **Processing happens on Cloudflare's global network.** Our code runs as Cloudflare Workers, which may execute in a data
centre outside the EU (for example the one closest to the sender's mail server or to your agent). Data is processed there
in transit, while a request or an email is handled, and not stored there.
- Cloudflare's Email Routing activity log (section 4.D) is kept by Cloudflare; Cloudflare does not document where.
- Cloudflare's Email Sending activity log and metrics (section 4.F) are kept by Cloudflare; Cloudflare does not document
where.
- Cloudflare's Workers AI, which converts attachments to text on request (section 6), has no EU jurisdiction setting.
Cloudflare does not document where it processes the data. The attachment is sent to it in transit; we store only the
converted text, in the EU.
- Cloudflare, Inc. is a US company. Because our code and these logs may be processed outside the EU, we treat this as a
transfer. It relies on the EU–US Data Privacy Framework (Commission Decision (EU) 2023/1795) and, as a fallback, on the
Standard Contractual Clauses in Cloudflare's Data Processing Addendum. You can ask us for a copy at privacy@replyz.io.
We do not sell personal data and do not share it for advertising.
| Recipient | What for | Data | In use at launch |
| --- | --- | --- | --- |
| Cloudflare, Inc. (and its EU affiliates) | Hosting, databases, storage, queues, receiving email, sending the emails your agent writes to you (Email Sending), running our code, and converting attachments to text on request (Workers AI) | All | Yes |
This list was last changed on 10 November 2026 (no new sub-processor: Workers AI is part of Cloudflare). We announce a new sub-processor at least **30 days** before it
starts handling your data, in the way described in Terms §15.2. If you do not agree, you can end the contract free of charge
before then.
We may also disclose data when the law requires it, for example to a court or a competent authority, and only what is
required. We check that each request has a legal basis and, where the law allows, tell the account owner.
**The AI model your agent uses.** When your agent reads your mail, the content goes to wherever your agent runs, including
the AI model provider you chose for it (for example the company behind the model your assistant uses). That provider is not
our sub-processor: you chose it, and its own terms and privacy policy apply.
## 9. If you sent an email to a Replyz address
- Your email is delivered to the AI agent that owns the address, and to its owner, through that agent or in their Replyz dashboard. That agent and its
owner decide what they do with it. The agent may pass your email to the AI model provider its owner uses.
- We store it for the periods in section 11 and run the automated checks in section 6 on it.
- If the address does not exist or cannot receive mail, or its owner has blocked you, your email is rejected with "mailbox unavailable" and we do not store
it; Cloudflare's activity log (section 4.D) still records the sender, recipient and subject.
- You have the rights in section 10, and you exercise them with us at privacy@replyz.io. We cannot undo what the agent or
its owner already did with your email (for example a copy the agent made).
- To report an agent that misuses its address, write to abuse@replyz.io.
## 10. Your rights
You have the right to access your data, to have it corrected or deleted, to restrict or object to its processing, and to
data portability, under the conditions of the GDPR.
- **Self-service:** the account owner can read every message through the API and download each one in its original form (the
complete message as it arrived, with attachments), so the mail can be taken to another provider. The owner can read the account's mail and the emails the agent sent them in the dashboard, and can delete the account on the manage page (replyz.io/manage). The agent can delete an inbox
through the API.
- **By email:** write to privacy@replyz.io. We answer within **one month**, which can be extended by two more months for complex
requests (we will tell you why). We may ask you to prove that you control the address or the account concerned.
- **Complaint:** you can complain to the Comissão Nacional de Proteção de Dados (CNPD), https://www.cnpd.pt, or to the data
protection authority of the country where you live or work.
## 11. How long we keep it
| Data | Kept for |
| --- | --- |
| Sandbox accounts (account, inboxes, mail) | Deleted **30 days after the sandbox expires** |
| Signup IP hash on the account | **30 days** |
| Messages of claimed accounts | **12 months** from receipt |
| Codes and links found in a message (extraction results) | **7 days** |
| Messages that look like one-time codes or sign-in links | Reduced after **7 days**: the original message file, its attachments, the extracted code and the message body are deleted; the subject, sender and date are kept until the message's normal retention ends |
| Events (the new-mail feed) | **30 days** |
| Activity timeline in the dashboard; sender domains of refused mail | **30 days** |
| Owner-view log (section 4.B) | **90 days** |
| Dashboard sessions (the session record) | Deleted within **2 days** after the session's 8-hour limit |
| Dashboard settings and sender lists | While the account exists; deleted with it |
| Text converted from an attachment (cache) | As long as the attachment: deleted with it, and after **7 days** for messages that look like one-time codes or sign-in links |
| Contacts (senders that passed authentication, per inbox) | While the inbox exists; deleted with the inbox or the account |
| Repeat-request protection records (idempotency) | **24 hours** |
| Abuse records | **12 months** |
| Log of the Director's reads of inbox content (who, when, which account, which report) | **12 months** |
| Founding Builder list entry (time, and the amount if given) | Until the account is deleted |
| Statements of reasons we sent about a restriction, and your objection | **12 months**, or longer while a complaint or dispute about it is open |
| Operational logs (internal IDs, status codes and timings only) | **14 days** |
| Mail of a deleted inbox | Hidden at once; deleted in the next daily run |
| Deleted account (at the owner's request) | Deleted within **7 days** of the request (address hashes kept) |
| Address hashes (so an address is never reissued) | **Permanently** (a keyed hash only, not the address) |
| Claim data of a claimed account (email address, terms version, time of the claim) | For the life of the account, then **3 years** as proof of the contract |
| Cloudflare's Email Routing activity log | Cloudflare's period, which Cloudflare does not document |
| Email from your agent to you (subject and text) | A copy for **12 months** from the moment your agent queues it, shown in the dashboard |
| Record of each summary, alert or warning email we sent you (kind, time, numbers, internal IDs) | **12 months** |
| Record of each notice we sent you (change of terms, new sub-processor) | **12 months** |
| Delivery events for those emails | Up to **1 day** |
| Address on the suppression list after a bounce or spam complaint | **2 years** after the last bounce or complaint, or sooner if you ask us at privacy@replyz.io |
| Cloudflare's Email Sending activity log / metrics | About **30 days** / **31 days** (Cloudflare's periods) |
These periods are applied by daily automated jobs. Deleted data may remain in our
providers' restore points for up to 30 days before it is gone.
## 12. Security
- API keys and address records are stored only as keyed hashes. The IP address is never stored.
- Every data store is in Cloudflare's EU jurisdiction (section 7).
- Mail, attachments and account data are stored only in Cloudflare R2, D1 and Durable Objects, which Cloudflare encrypts at
rest with AES-256. Our queue for incoming mail carries only IDs, a storage reference and the recipient address, never the
message itself; the queue for emails your agent sends you holds their subject and text until they are sent (section 4.F).
The API accepts only encrypted connections (HTTPS). Email reaches us
encrypted only if the sender's mail server uses TLS.
- Our own logs hold IDs only; no message content and no full addresses.
- Access to production systems is limited to the Director.
- If a personal data breach is likely to put you at risk, we tell the CNPD within 72 hours and, where required, tell you.
## 13. Age
Replyz is not meant for children. You must be at least 18 to claim an account.
## 14. Changes
We will update this notice when what we do changes, for example when replies are added (which will add new data and a new
provider). We tell account owners of important changes at least 30 days in advance, by email to the address they claimed
the account with, by a notice on replyz.io and in the account information the API gives their agent (`GET /v1/account`). Each version has a
date.