Speudio Privacy Policy

Version 3.7 Last updated: 17 October 2026

Language. This policy is written in English, which is the authoritative version. Swedish and

Greek translations are provided for convenience; if there is any difference between them, the

English version prevails. This policy is governed by Swedish law, and any disputes are subject to

the jurisdiction of the Swedish courts.

1. Who we are

Speudio (speudio.com) is operated by SK Creations, a sole proprietorship (Swedish enskild firma) registered in Sweden.

• Registered address: Hagvägen 41, 194 40 Upplands Väsby, Sweden
• Contact for all privacy matters: info@speudio.com

Because SK Creations is a sole proprietorship, its registration number is the owner's personal identity number and is therefore not published on this page. It is available on request via info@speudio.com, and is provided to business customers in the signature block of an executed Data Processing Agreement.

SK Creations has not appointed a Data Protection Officer, because none of the conditions in Article 37 GDPR that require one apply to our processing. The contact for every question, request, or concern about personal data is info@speudio.com.

For the data described in Section 3 under "account and security data", SK Creations is the data controller within the meaning of the GDPR.

2. Scope of this policy — two roles, one important distinction

Speudio is offered to businesses only, for use in the course of a business, trade, or profession (Section 1.2 of our Terms of Service). It connects to a mailbox you choose, with read-only access, imports your emails and attachments, builds contact profiles from them, and provides AI assistants that work over that data. Speudio never modifies, deletes, or sends mail from your mailbox.

That design means we process personal data in two distinct roles. This policy describes the first role in full; the second is governed by our Data Processing Agreement and appears in this policy only where needed to keep the distinction clear (for example, in the tables in Sections 3 and 7):

• Your account data — we are the controller. Information about you as our customer: your login details, plan, preferences, usage records, and security logs. This policy describes that processing.
• The content you import — you are the controller, and we are your processor. Your mailbox contains personal data about your contacts and customers (their emails, names, attachments, and the contact profiles and AI-derived data we build from them). Under the GDPR, you decide why and how that data is used; Speudio processes it only on your instructions, as your processor. That relationship is governed by our Data Processing Agreement (DPA) under Article 28 GDPR, not by this policy. If you are one of our customer's contacts and have questions about how a Speudio customer handles your data, the customer — as controller — is your first point of contact.

Where this policy says "imported content", it means everything Speudio derives from your mailbox: emails and their metadata, attachments (including images), CRM contact profiles, and AI-derived data such as captions, embeddings, summaries, and suggested questions.

3. What we collect, why, and on what legal basis

All of the data above comes from you: we create your account for you on request (you then receive a set-password invitation link), and the rest comes from your own use of the service. We do not collect personal data about you from third-party sources.

The consent-based processing we run is limited to two optional things you send us deliberately: the product feedback described above, and an answer you flag for AI improvement (Section 7). Neither is part of using the service, and declining both changes nothing else.

We send you service emails about your account: when a plan is set or changed, when a trial starts, is extended, is about to end or has ended, and when access ends; and each invoice we issue, with the PDF attached, to your invoice email address (or your login email if none is set). These are part of running your account, not marketing, and they are sent through our email provider, MXroute (Section 5). Beyond these, we send no marketing emails, and we use no advertising or behavioural tracking (see also Section 5 and our Cookie Notice). We never sell personal data, and we never send marketing email to your contacts.

We do not intentionally process special categories of personal data (Art 9 GDPR). Such data may appear incidentally inside a mailbox you import — for example, inside the text of an email someone sent you — but that sits in the imported-content layer, where you are the controller and the data stays under your control.

Two of the categories above can carry mailbox content out of that layer, and in both cases it is you who puts it there: a screenshot you attach to product feedback, and anything you quote into a support ticket. Section 12 names both as the only exceptions to our rule that no human at Speudio reads your imported mail, and Section 7 says how long each is kept.

4. Our legitimate interests

Where we rely on legitimate interests (Art 6(1)(f) GDPR), the interests are:

• Security and anti-abuse. Keeping accounts, the service, and the data within it safe requires retaining limited technical records — IP addresses, session data, and security events — for a short period (see Section 7). This includes a log of sign-in attempts. An attempt is recorded even when the email address entered belongs to no account, because credential-stuffing and brute-force attacks are aimed at addresses that do not exist just as often as at ones that do, and a log that omitted them would be blind to exactly the attacks it exists to catch. Without these records, we could not detect intrusions, investigate incidents, or stop abuse.
• Service improvement. Understanding how the service performs technically (errors, reliability) so we can fix and improve it.

We have weighed these interests against your rights and freedoms. The data involved is limited and technical, it is kept for a short period, it is not used to profile you or for marketing, and the processing is what a customer of a security-conscious service would reasonably expect. You can object to this processing at any time (see Section 8).

5. Recipients and sub-processors

We share personal data only with the following parties — and otherwise with no one, except where disclosure is required by law or by a binding order of a court or authority (see the end of this section). We use no analytics, advertising, or tracking vendors of any kind.

Eight honest notes on this list:

• Who at Speudio can read what. The rule is that no person here reads the mail Speudio imported for you, and it is kept by the way the software is built rather than by staff discipline: every query for email, attachments and contact profiles is scoped to the account that owns them. There are exactly two channels where a person does see content, and you open both yourself. The first is a screenshot you attach to product feedback: it is looked at by a person here, and because it is a picture of your screen, whatever was on the screen is in the picture. The second is a support ticket, which our support team reads, including any of your own email you quote into it. Support access is deliberately narrow even so, and this is worth being precise about: a member of the support team reading your ticket cannot open your mailbox, your attachments or your contact profiles from it, because those queries are scoped to your account and not theirs. What they see is the ticket you wrote.
• Your attachment files and the original copy of each message are stored on our own server in Sweden. Since 20 August 2026 the files attached to your imported mail (documents, images, PDFs) and the original formatted (HTML) copy of each message — the sender's own version, which is what you see when you open "view original" — are held on our own server in Sweden, in the same place as the database, rather than in outside object storage. Cloudflare previously held these files; it no longer does. The readable text we extract from each message stays in the same database, which is what the app and the AI assistant actually read. What protects this data is our control over the server, encrypted transport, and per-account scoping. Your mailbox credentials are encrypted (AES-256-GCM) and our backups are encrypted (restic); the server's own disk is not separately encrypted at rest today.
• OpenAI data handling. By default, data sent to the OpenAI API is not used to train OpenAI's models and is retained by OpenAI for at most around 30 days for abuse monitoring. Transfers to OpenAI take place under OpenAI's Data Processing Addendum (executed between SK Creations and OpenAI Ireland Ltd on 12 June 2026), which incorporates the EU Standard Contractual Clauses.
• OpenRouter, and why a company that only passes messages along is on this list. We are adding OpenRouter to this list before switching it on, rather than at the moment we do, so that you are told first. It changes the path an AI request takes, not the provider that answers it: the model is OpenAI's either way. What it changes for you is that one more company handles your content in transit, which is why it belongs here even though it keeps nothing we need it to keep. The reason we want it is unglamorous — OpenRouter tells us the exact amount each request cost, where working it out ourselves means multiplying token counts by a price list we maintain by hand. Three things are worth stating plainly. First, the pinning: OpenRouter can serve one model from several hosting companies, including Amazon and Microsoft, and a list of recipients that changes per request is not a list we could publish. We therefore restrict every request to OpenAI's own service and disable the automatic switching to any other, so the set of companies that see your content is the one named on this page. Second, retention: OpenRouter states that it does not log prompts or replies by default, and we additionally mark every request as excluded from any provider that would store it. Third, the limit of what we can promise: OpenRouter is a US company and we treat this as a transfer to a third country (Section 6), under a data-processing agreement and the EU Standard Contractual Clauses we execute with them before the route is used.
• Cohere data handling. Cohere's role grew in July 2026. It still orders search results by relevance, receiving only the search text and the text of up to 100 candidate documents (up to about 4,000 characters each, often most of an email). It is now also our embedding provider, which means the text of an imported email or document is sent to Cohere once, when it is added to the search index, so that a question asked in one language can find an answer written in another. Data we send to the Cohere API is not used to train Cohere's models (we have disabled this in Cohere's data controls) and is retained by Cohere for at most around 30 days for abuse monitoring.
• Backups. Nightly encrypted backups are kept in two places, both ours and both in Sweden: on our own server, and a second copy on separate hardware of ours kept apart from it. No outside company holds them. The attachment files that arrived with your mail and the original formatted copy of each message are deliberately left out, because they are a copy of your own mailbox and would be fetched from it again. Section 7 sets out what a backup holds, how long backups are kept, and what that means for you.
• Plans are paid by invoice, and Swish is switched off. Since 25 September 2026 a paid plan is paid by invoice, and no payment is taken inside the Service. The invoice details (Section 3) are kept in our accounting system in Sweden, outside the Service. Swish is built into the product and is kept ready rather than removed, but it is not offered and sends nothing. While it was in use, we connected to the Swish system directly with our own merchant certificate and sent it only the amount, the currency, a short message naming the plan, and a reference derived from your account identifier, never anything from your mailbox or your contacts. Getswish AB (org.nr 556913-7382), established in Sweden, operates the Swish system, and Lunar Bank A/S (org.nr 502089-5388), established in Denmark, is our bank and holds our Swish merchant agreement. A payment institution handles payment data under its own legal obligations rather than on our instructions, so we describe both as independent controllers of the payment execution, not as our sub-processors. Both are established within the EU/EEA. If Swish is switched on again, we will publish a new version of this policy, and tell you, before that happens.
• Stripe, the card integration, is switched off. Card payment is built into the product but is not offered to customers and processes nothing. It is paused and is kept ready rather than removed, which is why it stays on this list. No customer's name, email address or payment details have been sent to Stripe. If card payment is switched on later, it will run on Stripe's own hosted checkout page, so a card number would be entered on Stripe's page and never reach our servers, and Stripe would then receive your email address and the billing details you enter there. We will publish a new version of this policy, and tell you, before that happens.

We may also disclose personal data where required by law or by a binding order of a court or authority.

6. International transfers

The service itself runs in the EU. The service, its database, and all of your imported mail run on our own server in Sweden. Transactional email is sent through MXroute's mail servers in Germany; MXroute itself is a US company (see below). This is a self-hosted server rather than a certified third-party data centre, so we make no data-centre certification claim for it; Section 12 describes how we protect it.

The first exception is OpenAI (USA), our AI provider. When you use an AI feature, the relevant content is transmitted to the OpenAI API in the United States. That transfer is safeguarded under OpenAI's Data Processing Addendum incorporating the EU Standard Contractual Clauses (Art 46(2)(c) GDPR), executed with OpenAI Ireland Ltd on 12 June 2026, and OpenAI does not use API data for model training by default. The same transfer happens when our support team asks the assistant to summarise a support ticket you raised or to draft a reply to it: the text of that ticket is transmitted to the OpenAI API, under the same safeguard. You can ask us for more information about these safeguards, or for a copy of the relevant clauses, at info@speudio.com.

The second exception is Cohere (USA/Canada), our search provider. The text of your imported emails and documents is transmitted to the Cohere API to be converted into search embeddings, and when the AI searches, the search text and the text of up to 100 candidate documents (up to about 4,000 characters each, often most of an email) are transmitted to be ordered by relevance. That transfer is safeguarded under Cohere's data-processing terms incorporating the EU Standard Contractual Clauses (Art 46(2)(c) GDPR); Cohere retains API data for at most around 30 days and does not use it to train its models (disabled via Cohere's data controls).

The third is Cloudflare, a US-headquartered provider, which carries the traffic to speudio.com. Its network is global, so a request can pass through a Cloudflare data centre near the person using the site, which may be outside the EU. Cloudflare passes the traffic on and does not keep it. We treat this as a transfer to a third country, safeguarded under Cloudflare's Data Processing Addendum incorporating the EU Standard Contractual Clauses (Art 46(2)(c) GDPR); Cloudflare is also certified under the EU-U.S. Data Privacy Framework. Cloudflare holds none of our backups and none of your files.

The fourth is OpenRouter, Inc. (USA), and it is the one exception on this list that carries no new kind of data. Speudio can reach the OpenAI models it uses either directly or through OpenRouter. On the second route the same content is transmitted to the same AI provider, with OpenRouter in the middle, so what changes is not what leaves the EU but how many companies handle it on the way. OpenRouter, Inc. is established at 169 Madison Avenue, New York, NY 10016, United States. That transfer is safeguarded under OpenRouter's data-processing agreement incorporating the EU Standard Contractual Clauses (Art 46(2)(c) GDPR), which we execute before the route is used; requests are restricted to OpenAI's own service so that no further hosting company becomes a recipient, and each request is marked as excluded from any provider that would retain it.

The fifth is MXroute LLC (USA), which sends our service emails (Section 5) from its mail servers in Germany. Because MXroute is a US company, we treat this as a transfer to a third country. MXroute does not offer a data-processing agreement or the EU Standard Contractual Clauses, so no such safeguard is in place for it. We keep what it handles small: it carries only the emails about your account listed in Section 3, including invoices, and never your imported mail or anything from your mailbox.

Paying for a plan stays inside the EU. Plans are paid by invoice, which we issue and keep in Sweden. Swish, which is switched off, is operated by Getswish AB (org.nr 556913-7382) in Sweden, and Lunar Bank A/S (org.nr 502089-5388), our bank and Swish acquirer, is established in Denmark. All of these are inside the EU/EEA, so paying for a plan adds no third-country transfer and needs no safeguard such as the Standard Contractual Clauses.

Stripe is not a transfer of any kind today, because the card integration is switched off and no customer data has been sent to it. If it is switched on, Stripe Payments Europe Ltd (Ireland) will be the contracting entity, with onward transfer to Stripe, Inc. (USA) under Stripe's data-processing terms and the EU Standard Contractual Clauses. We will update this section before that happens.

Beyond these, we transfer personal data to no other recipient outside the EU. One practical note: if the mailbox you choose to connect is hosted outside the EU, that is part of your own relationship with your mail provider. The provider is the data source you bring to Speudio, not a recipient we send data to (see Section 5).

7. How long we keep data

Backups. Encrypted backups are made every night and kept on a rotating schedule of 14 daily, 8 weekly and 12 monthly snapshots, both on our own server and in the second copy on separate hardware of ours in Sweden. There is no yearly tier. Two parts of your imported mail are left out on purpose: the attachment files that arrived with your messages, and the original formatted copy of each message (what you see under "view original"). They are a copy of your own mailbox, so if we ever had to rebuild the service they would be fetched from your mail server again, which only works while your mail server still has them. Everything else is in the backup, including each message's sender, recipients, subject, date and readable text, your contacts and the profile details, notes, tags and custom fields you keep on them, your settings and mailbox connections, your AI conversations, files you uploaded yourself, your support tickets, your payment and receipt records, and the invoice copies stored in your account. When you delete data, or your whole account, it is removed from the live service immediately, as described above, and whatever part of it was in a backup ages out as those snapshots rotate. Backups exist solely for disaster recovery: they are never used to "un-delete" data, and their contents can only be read with a decryption key that only we hold.

8. Your rights

Under Articles 15–22 GDPR you have the rights below. Wherever a self-service path exists in the app, it is listed — you can exercise the right yourself, instantly, without asking us. For everything else, email info@speudio.com; we respond within one month (Art 12(3) GDPR).

Exercising these rights is free of charge. If you are a contact of one of our customers (i.e. your data was imported by a Speudio user from their mailbox), please direct your request to that customer, who is the controller of that data — we will assist them as their processor under the DPA.

9. Complaints

If you believe our processing of your personal data infringes the GDPR, you have the right to lodge a complaint with a supervisory authority. Our supervisory authority is the Swedish Authority for Privacy Protection:

IMY — Integritetsskyddsmyndigheten, www.imy.se

You may also complain to the supervisory authority of the EU member state where you live or work. We would, of course, appreciate the chance to address your concern first at info@speudio.com — but you are never required to contact us before going to the authority.

10. Is providing your data required?

There is no statutory obligation to provide us with personal data. Contractually, however, the account data in Section 3 (login email, password, and similar) is necessary to create and operate your account — without it, we simply cannot provide the service. Connecting a mailbox is your choice; the service only becomes useful once you do, but which mailbox you connect, and what it contains, is entirely up to you.

11. No automated decision-making

We do not carry out any automated decision-making, including profiling, that produces legal effects concerning you or similarly significantly affects you (Art 22 GDPR). Speudio's AI features only assist you — they summarise, caption, suggest, and answer questions over your own data; every decision remains yours.

12. How we protect your data

• Self-hosted in the EU (Sweden). The service, its database, and all imported mail run on our own server in Sweden. It is not a certified third-party data centre, so we make no such certification claim; the measures below are how we protect it.
• Photo search facts are worked out on our own server, and we never recognise faces. To make photo search work, Speudio can work out a few facts about each photo in your imported mail: a picture vector (a list of numbers that describes what the photo looks like), how many faces and how many people it shows, what kinds of things it shows (for example a person, a book or a chair), and its main colours. This happens on our own server in Sweden, with software that runs there; the photos and these facts are not sent to anyone. We count faces and we never use any of these facts to recognise, identify or compare people. For faces we keep only how many there are, how sure the count is, and how large each one appears in the photo. The facts are deleted together with the photo.
• Text in scanned pages and photos of documents is read on our own server. Where this is switched on for your mailbox: when a file is a scanned page or a photo of a document, our server reads the printed text in it so that you can search for it. This happens on our own server in Sweden, with software that runs there; the file is not sent to anyone for it. The text is kept with the file and is deleted together with it. After that it is used like the text of any other document you have imported, for example when you search or ask the assistant about the file (Section 5).
• Encrypted transport (TLS) for all connections to the service.
• Passwords stored only as bcrypt hashes — never in plain text.
• Per-user tenancy isolation enforced server-side: your data is partitioned from every other customer's at the application layer on every request.
• Read-only mailbox access, whichever provider you connect. Speudio never modifies, deletes, moves, marks, or sends mail in your mailbox. How that is enforced differs by provider, and the detail is set out immediately below this list.
• Attachment files stored per-account: each file is written under a key belonging to your account alone, so deleting your account removes every one of them (Section 7).
• Recovery codes for secure account recovery.
• No human reading of your email content, with two exceptions you open yourself. AI features process your content programmatically, and our staff do not read the emails Speudio imported for you. The two exceptions are both things you send us on purpose, and we would rather name them than let the promise read wider than it is. A screenshot you attach to product feedback is looked at by a person here; it is a picture of your screen, so whatever was on the screen is in the picture, and it is deleted within 90 days (Section 7). A support ticket is read by our support team, including any of your own email you quote into it while explaining the problem. Neither is required: a written note with no screenshot is a complete feedback submission, and a ticket only contains mail if you paste it in. Support access is narrow in any case, and it is the software that keeps it narrow rather than a rule staff are asked to follow: every query for email, attachments and contact profiles is scoped to the account that owns them, so a member of the support team reading your ticket cannot reach your mailbox or your contacts from it.
• No training of AI models on customer data, with one exception you control: an answer you explicitly flag and confirm is stored so we can improve Speudio's own assistant. It is never shared with a model provider for training, and it is deleted within 90 days.

How read-only access is enforced, by provider. Outlook, Hotmail and Microsoft 365 mailboxes connect through Microsoft Graph, where we request only Mail.Read, plus the permission needed to keep the connection alive and the minimum sign-in information needed to label the account. We deliberately do not request Microsoft's read-write, send, or user-profile permissions, so Microsoft itself enforces the restriction.

Gmail and every other provider are read over IMAP, and Speudio opens every folder in IMAP's read-only mode, so the connection cannot carry a write, move, flag, or delete command. Gmail is connected with an app password that you create in your own Google account. For Gmail there is one point of honesty owed: an app password is not limited to reading at Google's end, so the restriction is enforced by the way Speudio connects rather than by the password itself, and it applies without exception. You can revoke the app password in your Google account at any time, which cuts Speudio off from that mailbox at once.

13. Google user data (Calendar and Tasks)

Google Calendar is not offered in Speudio at the moment, and no Calendar or Tasks data is accessed. This section describes how it will work once it is switched on, and we will tell you before that happens.

Connecting Google Calendar to Speudio is optional. If you choose to connect it, this section explains exactly what Google data Speudio accesses and how we handle it. You can disconnect at any time (Settings → Mailbox → Google Calendar → Disconnect).

What we access. When you connect, you grant Speudio these Google permissions:

• See and edit events on your calendars (calendar.events) — so Speudio can show your events and create or change the events you ask it to create or change.
• See your calendars (calendar.readonly) — so Speudio can list your calendars and let you choose which ones appear in the app. This is a narrower, read-only permission than the full calendar permission; we use it to list and mirror the calendars you select.
• Manage your tasks (tasks) — so Speudio can show your Google Tasks lists and let you add, edit, complete and remove tasks.
• Your Google account's email address (openid, userinfo.email) — read once at connection time, only to confirm and label which Google account you connected.

Why we need it. These permissions let Speudio show your calendars, events and tasks in one place alongside your inbox and contacts, and let you (or, only with your explicit confirmation, the AI assistant) create and update events and tasks without leaving Speudio.

Where it is stored, and for how long. To make your calendar and tasks fast to use, Speudio keeps a mirror copy on our own server in Sweden (the same hosting described in Sections 5 and 12): your calendars and their details, your events, your task lists and tasks, the connected Google account's email address, and your Google authorisation tokens (encrypted at rest). This mirror exists only while your connection is active. When you disconnect Google Calendar, Speudio deletes the connection together with its mirrored calendars, events, task lists, tasks and stored Google tokens from our live systems. Deleting your Speudio account removes all of this as well (Section 7). Encrypted backups age out on the rotation schedule in Section 7.

How we use it. We use your Google Calendar and Tasks data only to provide the calendar and tasks features inside Speudio — showing your events and tasks, and making the changes you ask for. If you use Speudio's AI assistant to ask about your calendar or tasks (for example "what's on Friday?" or "remind me to call Anna"), the relevant events or tasks are processed by our AI provider to produce the answer, under the same safeguards described in Sections 5 and 6. We do not add your calendar or tasks to our search index, and we never use them to train any AI model. We do not sell your Google data, and we do not use it for advertising. We do not transfer it to anyone except the sub-processors already listed in Section 5, and only for the purposes described there.

Limited Use. Speudio's use and transfer of information received from Google APIs to any other app will adhere to the Google API Services User Data Policy, including the Limited Use requirements.

14. Children

Speudio is not directed at children. You must be at least 18 years old (or the age of majority where you live) to create an account, and we do not knowingly process children's personal data as a controller.

15. Changes to this policy

This policy is versioned (see the version number and "last updated" date at the top). We update it whenever the facts change — for example, when a sub-processor is added or removed (Section 5), or when retention or data categories change. Material changes will be notified to you in-app or by email before they take effect. Earlier versions are available on request.

In version 1.2 (19 July 2026) we added Section 13, describing how Speudio accesses, stores and uses your Google Calendar and Tasks data when you connect Google Calendar.

In version 1.3 (21 July 2026) we described the record we keep of sign-in attempts, password changes and recovery-code use, including attempts made against email addresses that belong to no account, and its 90-day retention (Sections 3, 4 and 7).

Version 1.4 (22 July 2026) is a correction, and the largest of the changes concerns where your files are kept. Since 20 June 2026 your attachment files have been stored by Cloudflare rather than on our own server, while this policy still described Cloudflare only as the holder of encrypted backups. Section 5 now says plainly what is stored there and how it is protected, Section 6 states what we do and do not know about where those files sit, and Section 12 explains how read-only mailbox access is enforced for each provider, including the Microsoft Graph connection for Outlook accounts. This version also records that Cohere became our embedding provider in July 2026, that Cloudflare carries our network traffic, and that Stripe will become a recipient when online payment launches. It also corrects the Free-plan email window from 2,000 to the 200 most recent emails, which is the limit the service has applied since 19 July 2026.

Version 1.5 (25 July 2026) records that paid plans are now paid for with Swish. Section 5 adds Getswish AB, which operates the Swish system, and Lunar, our bank and Swish acquirer, and explains exactly what is sent to Swish and what comes back, including the payer's Swish number. The Stripe entry is corrected: the card integration is built into the product but is switched off and processes nothing, so it is no longer described as merely "not yet in use". Section 3 adds the payment and billing records we keep, Section 6 states that paying for a plan involves no transfer outside the EU, and Section 7 explains that the accounting records behind a purchase are kept in our accounting system for the period Swedish bookkeeping law requires.

Version 1.6 (25 July 2026) identifies the two payment parties precisely. Sections 5 and 6 now give the full legal name, registration number and country of establishment of each: Getswish AB, org.nr 556913-7382, established in Sweden, and Lunar Bank A/S, org.nr 502089-5388, established in Denmark. Both are inside the EU/EEA, so paying for a plan involves no third-country transfer and needs no safeguard such as the Standard Contractual Clauses. No new recipient was added and nothing about what is sent, received or retained has changed.

Version 1.7 (25 July 2026) extends how long we keep the answers you flag for AI improvement (Section 7), from 30 days to 90 days (about three months), so we have time to review the flagged answers and improve the assistant. Nothing else about this data changes: it stays optional and consent-based, no human reads the content, it is never used to train third-party models, and it is deleted automatically once the retention period ends.

Version 1.8 (26 July 2026) rewrites the retention of imported content in Section 7. Every plan now states the number of emails it holds (Free 5,000, Solo 50,000, Pro 150,000, Business 400,000), where the paid plans previously stated no limit and the Free plan a moving window of its 200 most recent emails. The mechanism changes with it, in your favour: imported email is no longer removed to make room for newer email, and a full plan stops importing new email instead. Nothing else about how this content is handled changes, and no recipient was added or removed.

Version 1.9 (26 July 2026) lowers three of the four email figures given in Section 7. They were introduced in version 1.8 earlier the same day and were not in force before it. Solo now holds 25,000 emails in place of 50,000, Pro 60,000 in place of 150,000, and Business 100,000 in place of 400,000; the Free plan's 5,000 is unchanged. Nothing else changes: the figure is still a ceiling rather than a moving window, imported email is still never removed to make room for newer email, and a full plan still stops importing instead. No recipient, sub-processor, transfer or category of data was added or removed.

Version 2.1 (28 July 2026) makes two promises more precise rather than changing what happens. Our marketing pages said we never train AI models on your content, without naming the one exception this policy already described: an answer you deliberately flag and confirm. That exception is now stated everywhere the claim appears. Separately, where we previously said no human reads a flagged answer, we now describe what the system actually enforces, which is that flagged content never appears in our admin screens (they show only how many flags exist). No data is processed differently as a result of this version.

Version 2.2 (29 July 2026) adds two categories of data this policy had never described, and tightens one deletion promise. Product feedback and support tickets are now in Section 3 with their purposes and legal bases, and in Section 7 with their retention. Both were already being collected; neither appeared anywhere in this document, and the feedback tool had no retention period at all, so a screenshot was kept until someone removed it by hand. From today a nightly job deletes feedback, and any screenshot attached to it, within 90 days. Section 5 adds a plain statement of who at Speudio can read what, and Section 12 keeps the promise that no human reads your imported email while naming the two exceptions that already existed: a screenshot you attach to feedback, and anything you quote into a support ticket. Section 5 and Section 6 now also state that a support ticket's text goes to OpenAI when our support team asks the assistant to summarise it or draft a reply, under the safeguard already described there. Section 3's sentence saying we run no consent-based processing was wrong the moment flagged answers existed and is corrected. Apart from the new 90-day deletion, nothing about how any of this data is handled has changed; what changed is that the document now says so. Section 7 also describes account deletion more precisely: your files are deleted before your account record, and if the files cannot be deleted nothing is deleted and you are asked to retry, which replaces the earlier description of a single transaction. No recipient or sub-processor was added or removed.

Versions 2.3 and 2.4 (30 July 2026) describe an export that now does what this policy's access and portability rows always claimed. The export previously covered 8 of the 45 tables we hold about an account and contained no files at all; version 2.3 records the rebuilt version, which covers every table and includes your files. Version 2.4, the same day, records one refinement made before either reached anyone: the email Speudio mirrors from your mailbox is left out of the download unless you ask for it, because that same mail is still in your own mailbox and including it makes the file very large. Ticking Include my email and its attachments produces the complete archive, and the archive itself names what it contains, what you chose not to download, and the few categories we cannot release, with a reason for each. No recipient, sub-processor, transfer or category of data was added or removed, and nothing is retained differently.

Version 2.5 (2 August 2026) adds one sub-processor, OpenRouter, Inc. (USA), to Sections 5 and 6, before it is switched on rather than at the moment it is. It is a router: Speudio can reach the OpenAI models it already uses either directly or through OpenRouter, and on the second route the same content goes to the same AI provider with one more company handling it in transit. No new category of data is collected, nothing is retained for longer, and no new AI provider is introduced. Because OpenRouter can serve one model from several hosting companies, we restrict every request to OpenAI's own service and switch off automatic re-routing, so the recipients are the ones this policy names. The reason for the change is measurement: OpenRouter reports the exact cost of each request, where we otherwise estimate it from a price list we maintain by hand.

Version 2.6 (5 August 2026) moves one more part of your mail out of our own database and into Cloudflare's R2 object storage, where your attachment files have been kept since 20 June 2026. The part that moves is the original formatted (HTML) copy of each imported message, which is what you see when you open "view original". The readable text of every message, its subject, and who it was from and to all stay in the database on our hosting provider's servers, and Sections 5 and 6 now describe the split. This adds no recipient and no sub-processor: Cloudflare already held your files and already carried our network traffic, and it is named in this policy for both. Nothing about what we collect, why, or how long we keep it changes. Deleting your account still erases this content along with everything else, because the erase covers that storage area in the same way it covers your files.

Version 2.0 (27 July 2026) removes a retention rule from Section 7 rather than adding one. Moving to a plan that holds fewer emails than the account already has previously locked the oldest ones above the new number and deleted them from the service after about two months. It now removes, hides and deletes nothing: everything already imported stays, the account simply sits above its limit, and the service stops importing new email until it is back under. No recipient, sub-processor, transfer or category of data was added or removed.

Version 2.7 (8 August 2026) removes the email count from the retention of imported content in Section 7. Every plan previously held a stated number of emails (Free 5,000, Solo 25,000, Pro 60,000, Business 100,000). That figure is gone, and the number of emails an account holds is no longer limited. Storage is now the only limit, and the row states it: Free 3 GB, Solo 15 GB, Pro 30 GB, Business 50 GB, covering your imported email and the files that arrived with it, counted together and measured as your own mail provider measures them. What we derive from your mail in order to run the service — the search index, the embeddings, the text we extract and the captions we generate — is not counted against it. Nothing else about how imported content is retained changes: the figure is still a ceiling rather than a moving window, imported email is still never removed to make room for newer email, and a full plan still stops importing instead. No recipient, sub-processor, transfer or category of data was added or removed, and nothing is kept for longer.

Version 2.8 (8 August 2026) says what our backups contain, in Sections 5 and 7. Two parts of your imported mail are deliberately left out of them: the attachment files that arrived with your messages, and the original formatted copy of each message. Those are a copy of a mailbox you already own, so we no longer write them to a second place; if we ever had to rebuild the service they would be fetched from your own mail server again. Everything else, including each message's sender, recipients, subject, date and readable text, stays in the backup. This is a reduction in how much of your content we hold rather than an addition, and no recipient, sub-processor, transfer or category of data was added or removed. It also has a limitation we would rather you read here than discover later: a rebuild can only bring those two parts back if your own mail server still has the messages, so Speudio is a companion to your mailbox and not an archive that replaces it.

Version 2.9 (20 August 2026) records that Speudio has moved to our own server in Sweden and changes what each outside provider does. The service, the database, and all of your imported mail — including the attachment files and the original formatted copy of each message — now run on our own self-hosted server in Sweden, where previously the service was hosted by one.com and those files and copies were held in Cloudflare's object storage. Three consequences run through Sections 5, 6, 7, 12 and 13. First, one.com is now only our transactional-email provider; it no longer hosts the service or the database, and no data-centre certification is claimed for our own server. Second, Cloudflare's role is reduced to holding our encrypted off-site backup: it no longer stores your files or message copies, and no longer carries the traffic to speudio.com. Third, because your attachment files and message copies now live only on our own server, the backups include them, so that this content keeps a second, encrypted copy; the backup schedule is set out in Section 7 (7 daily, 4 weekly and 6 monthly snapshots off-site, with a more frequent copy on our own server, and no yearly tier). No new recipient or sub-processor is added, and nothing about what we collect, why, or how long we keep it changes.

Version 3.0 (25 September 2026) records how plans are now paid for. Paid plans are paid by invoice, and Swish is switched off: it is built in and kept ready, like the card integration, but it is not offered and sends nothing. Section 3 now describes the invoicing records we keep (the business's name, organisation number, billing address, invoice email, and each invoice) in place of the Swish payment records, which are kept only for payments already made with Swish. Sections 5 and 6 mark Getswish AB and Lunar Bank as switched off. The account data in Section 3 now names the free trial, which replaces the Free plan for new accounts, and the storage figures in Section 7 drop the Free plan's. No recipient or sub-processor is added, and nothing is sent anywhere new.

Version 3.1 (25 September 2026) lowers the storage figures in Section 7: Solo 10 GB, Pro 15 GB, Business 30 GB (from 15, 30 and 50 GB); a trial has the Pro figure. Nothing already imported is removed, and no data is collected, kept or sent any differently.

Version 3.2 (25 September 2026) corrects Section 2, which still called Speudio a service for businesses and private users: it is offered to businesses only, as our Terms of Service say. Section 3 now says that a copy of each invoice we issue is stored as a PDF in your account and deleted together with it, and lists the service emails we send about your account (plan changes, trial start and end, invoices). Section 5 says these emails, including each invoice, are sent through one.com, which was already our email provider. Section 7 adds these invoice copies to the list of what a backup holds. No recipient or sub-processor is added.

Version 3.3 (27 September 2026) corrects several descriptions that no longer matched how the service runs. Cloudflare does carry the traffic to speudio.com, through a secure tunnel to our server, and it no longer holds any backup; Sections 5 and 6 now say so. Our backups are kept in two places, both ours and both in Sweden, with 14 daily, 8 weekly and 12 monthly snapshots, and they leave out the attachment files and original copies of your mail, which come from your own mailbox; Sections 5 and 7 now say so. Our service emails are sent through MXroute, not one.com as the previous version said; Sections 3, 5 and 6 now say so, including that MXroute is a US company that offers no data-processing agreement. Section 7 now says that an account whose trial ended without a plan is deleted 90 days after the trial ended, with an email 7 days and 1 day before. Section 3 says that accounts are created on request, as public sign-up is closed. Section 7 gives the new storage figures: Solo 3 GB, Pro 6 GB, Business 25 GB, and a trial 6 GB; nothing already imported is removed. Section 12 says that Gmail is connected with an app password, and Section 13 that Google Calendar is not offered at the moment.

Version 3.4 (8 October 2026) describes two transfers more exactly. Sections 5 and 6 say that for relevance ordering Cohere receives the text of up to 100 candidate documents, up to about 4,000 characters each, rather than "short excerpts" as earlier versions said. Section 5 also says what a web search sends: only the query the assistant writes, with email addresses, links and long numbers removed, which OpenAI's web-search tool passes to its search partners. No recipient or sub-processor is added.

Version 3.4 (8 October 2026) describes the short form on the sign-up and trial-ended pages, where you can ask for a trial or a plan with your name, your business's name and your email address (Section 3), how long we keep such a request (Section 7), and that the confirmation email goes out through MXroute like our other service emails (Section 5). Section 3 also says that the priority of a support ticket is now set by our support team rather than chosen by you. No recipient or sub-processor is added.

Version 3.5 (15 October 2026) describes the facts photo search can use (Sections 3 and 12): a picture vector, the number of faces and people, the kinds of things shown, and the main colours of each photo in your imported mail, worked out on our own server in Sweden. Faces are counted, and the facts are never used to recognise, identify or compare people. No recipient or sub-processor is added, and nothing leaves our server for this.

Version 3.6 (16 October 2026) says that our own server can read the printed text in scanned pages and photos of documents so that you can search for it (Sections 3 and 12). The text is kept with the file and deleted with it. No recipient or sub-processor is added, and nothing leaves our server for the reading itself.

Version 3.7 (17 October 2026) adds contact pictures to Section 3: a picture you choose for a contact, and, where you leave "Add company logos" switched on, the logo our own server fetches once from the public website of the company in a contact's email address (never for Gmail, Outlook or other personal mailboxes). That request goes only to the company's own website and carries nothing about you or your contacts, so no other service learns who your contacts are. The picture is kept with the contact and deleted with it. No recipient or sub-processor is added.

SK Creations (sole proprietorship registered in Sweden) Hagvägen 41, 194 40 Upplands Väsby, Sweden info@speudio.com