What we hold, and what we don’t.

1. Who is responsible for what

There are two kinds of data here and the law treats them differently.

If you self-host Minimal RM on your own infrastructure, we hold nothing at all — the sections below about our hosting, our subprocessors and our access do not apply to you.

2. What we collect

Things you create

CRM and IRM records — contacts, companies, deals, tasks, investors, missions, playbooks, notes, files and the full history of changes to them. Anything you import, whether from a CSV, a spreadsheet, a PDF, a screenshot or pasted text.

Things connectors bring in

If you connect Slack, Notion, a WhatsApp export or your own WhatsApp as a linked device, we pull the messages and content those scopes cover and match them to your records.

Connecting Google is one grant covering several things, and we name them because “Gmail” would understate it: we read and organise your Gmail inbox and sent mail (labels, stars, archiving, drafts), we send messages you have reviewed and clicked Send on, we read every calendar on the account to show meetings against the right people and, with a permission you grant separately, create, edit, move and delete events on it too, and we read a Google Sheet when you import from one. The connection card inside the product lists each of these against what you actually approved, with an upgrade offered for anything you have not, so you can see at a glance which ones are live. You can revoke any of it at Google, or disconnect, at any time.

Meetings, if you ask for it

You can paste or upload a transcript at any time. Separately, an admin can enable our meeting bot. The bot never joins a meeting on its own: someone has to invite it per meeting and confirm that participants have been informed. It appears in the participant list as “Minimal AI” and announces in the meeting chat that it is recording. Recording can be switched off, and the bot can be removed at any point.

A transcript is split into turns and each speaker is matched to a person in the workspace where we can tell — which means a record of what a named person said, and we treat it that way. The matching is evidence-based and refuses to guess: the meeting’s own invitee list first, then people already in the workspace by name or email, and anything ambiguous (two people sharing a name, an unrecognised voice) is left explicitly unidentified rather than attributed to somebody who might not have said it. A person in the workspace can correct or clear any attribution, and doing so moves the record of participation with it. The talk-time figures beside a transcript are arithmetic — seconds, turns, words, question marks — and never an AI judgement about how somebody behaved on a call. Transcripts, their turns and the “this voice is this person” mappings are all in your data export, and deleting a person removes their mappings and clears their name off every turn.

A call can also be read for what it implies about each person on it — a change to their record, a segmentation trait, a personal detail, a promise made on either side. Everything it produces is a proposal: nothing is written to anybody’s record until a person in the workspace reviews it and clicks apply, and every proposal must quote the transcript line it came from — anything that cannot be tied to a line is discarded rather than guessed at. A number somebody says out loud is stored as quoted and unverified, never as a checked figure. A record that has personal notes turned off is excluded before the analysis runs, not filtered afterwards. A follow-up email drafted from a call is a draft: it opens in the composer for a person to read and send, and nothing in the product can send it for them. Call proposals and the commitments tracked from a call are both in your data export, and deleting a person removes every one of them.

Personal details about the people in your CRM

The product can keep human details that make an interaction warm — a birthday, a new job, a holiday someone mentioned. Three deliberate limits: they are only ever what a source you already ingested actually said, never inferred; every one arrives as a proposal that a person has to accept; and health, religious, political, sexual and racial characteristics are excluded outright — there is no category for them and the model is instructed never to record them. You can turn the whole thing off per record, which purges what is already held.

Notes you keep on a dashboard

A dashboard can hold a notepad. What you type in one is stored against you personally, not against the dashboard, so a colleague opening the same dashboard sees their own empty notepad rather than your text — and a dashboard you share carries the layout only, never anything anybody wrote in it. Your notes are deleted when the widget or the dashboard is, and they are included when your organisation exports its data.

Sharing a widget or a dashboard with colleagues

You can publish a widget, or a whole dashboard, to your organisation. What is published is the recipe — which figures it shows, how they are grouped, how it is laid out — and never the figures themselves. Nobody inherits your numbers: when a colleague adds one, it is re-read from scratch against the records their own role is allowed to see, so somebody without access to a part of the workspace sees a tile that says so rather than your data. A published notepad arrives as a blank pad. Publishing is restricted to people your admin has given the permission to, and every publish, rename and removal is written to your organisation’s activity log.

People you send a link to

When a customer of ours shares a deck or a document with someone through a link, the page that opens says so: it records that it was opened, and how long was spent on each page, and the person who sent it can see that. Those views are tied to a record in their workspace where the link was created for a named investor, or where the address typed at the door matches someone they already hold — and where neither is true the view stays marked as unattributed rather than being matched to a likely-looking person. A customer can attach a view to a record themselves, which is a decision they make and we log, never one we infer. We never store a full IP address for a link view, and none of this is used for advertising or shared with anyone.

Alongside the opens and the time on each page we record, for each visit, the kind of device (phone, tablet or desktop), the operating system and browser, the size of the browser window, and the country. The country is the only location: it comes from the network the request arrived on, and no IP address is stored in any form — not the full address and not a truncated one. The browser’s user-agent string is read to work out the device and browser and is then discarded rather than kept. There is no tracking of any kind beyond the shared page itself. The page says all of this to the person reading it, in the same words, before they scroll. Where a customer asks our AI to interpret a viewing pattern, it is given only these recorded figures — it is not given anything else about the reader, and it cannot see anything we have not listed here.

How you use the product

While you are signed in we record a coarse, country-level location so we can see roughly where the product is used. It is derived only from your browser’s time-zone setting — never your IP address, never coordinates — stored as a country-level point and deleted after 30 days. We also record which AI features ran and what they cost, so we can bill accurately and show you where credits went.

Email engagement

Email you send through Minimal RM can report opens and clicks, using a tracking pixel and wrapped links. Your admin can switch this off for the whole workspace. Replies, bounces and unsubscribes are always detected regardless, because they are how we keep you from emailing someone who asked you to stop.

Open tracking is a one-pixel image, and it is not a reliable count of anything: mail privacy features and image proxies fetch it whether or not a person read the message, and someone who blocks images never fetches it at all. The product reports opens as a floor with that caveat printed beside them rather than presenting them as fact.

Email preferences and consent

Every campaign email sent from a workspace — an investor update, an outreach campaign, a sequence step — carries a one-click unsubscribe and a link to a preference page, where the recipient can choose which kinds of mail they want, or stop all of it. That page is reached only through a signed, expiring link sent to that person; it never asks for an address, never confirms whether one is on file, and shows nothing about the workspace beyond the sender's own name.

A one-to-one message — a reply, or a note somebody wrote to one person — carries neither, because it is not campaign mail: no unsubscribe footer is added to it and no List-Unsubscribe header is set. Where the workspace has engagement tracking switched on, a personal message still carries the open pixel and rewritten links described above; that is the tracking setting, and it is separate from the unsubscribe. An unsubscribe applies to the workspace as a whole, so once somebody has asked a workspace to stop, that is honoured for every kind of email including a personal one.

When someone changes their preferences we record what they chose, when, which email the link came from, a coarse IP and browser family, and any reason they typed (always optional). That record is their evidence as much as the workspace's: it is included in a data export, and it is erased when the record it belongs to is deleted. The resulting preference itself — and any unsubscribe — is not erased with the record, deliberately: if it were, deleting and re-importing a contact would be a way to undo somebody's decision to stop hearing from you. Preferences are enforced when a message is queued and again when it is actually sent, so a choice made in the seconds after someone clicks Send is still honoured.

The mail client, and letting someone else act in your mailbox or calendar

Once Google is connected, Minimal RM keeps its own working copy of that mailbox — message text, subject and who is on each thread — so the mail client can be fast and searchable without asking Gmail on every click. It is a rolling window, 24 months by default: the copy is what makes the client work, never a second archive kept for its own sake, and an admin can widen or narrow the window (1 to 36 months) in Admin → Data. Purging what falls outside it never loses the mail itself — opening or searching an older thread simply fetches it from Gmail again and re-caches it, exactly as reaching past the window always has. An unsent draft is never purged, however old the thread under it is. Attachment FILES stay in Gmail until somebody opens one; only then is that one file pulled into the workspace's own storage.

This copy belongs to the person who connected the mailbox, and by default nobody else can read it. They can name a specific colleague and grant them read, draft, send or label access — never wider than what they hold themselves, always with an honest sentence of exactly what it allows, and never Google-native delegation: it is Minimal RM's own audited grant, sitting on top of the owner's Google connection. It can carry an expiry, and pausing or revoking it takes effect on the very next request, with nothing left to catch up or expire on its own. Every read made under a grant, and every write, is logged against both people — the colleague who acted and the mailbox owner — and both are told the moment access is granted, paused, resumed or revoked. The recipient of an email sent this way sees only the mailbox owner's name; there is no “on behalf of” anywhere in the message itself.

The calendar works the same way: full read and write against every calendar on a connected account, with Google remaining the source of truth for every event, and sharing yours with a colleague uses the identical grant — named, scoped, optionally time-limited, instantly revocable, and every change they make on it recorded against both of you. Deleting a contact or investor removes the cached mail threads filed against them or carrying their address, and says how many so restoring the record is never mistaken for restoring the mail too; deleting the organisation removes all of it, cache, grants and calendars included.

Connected mailboxes (Pulled info)

A connected mailbox belongs to the person who connected it, not to the workspace. Each connection carries its own sharing setting, and the default is record-matched only: a message becomes visible to colleagues when this app can match it to a contact, company or investor already in the workspace. Everything else stays readable by its owner alone — it still appears on their own Pulled info page, marked so, and nobody else's search, export, report or AI answer can reach it.

The owner can narrow that further to private to me, which shares nothing at all, or widen it to everything in the sync window — a deliberate choice that requires typing a confirmation, is written to the workspace's activity log, and re-checks the messages already pulled, reporting how many changed either way. Sharing can also be limited to named labels or folders. Matching a private message to a record shares it; removing that match takes it back; both are logged.

An organisation-wide data export therefore carries the workspace-visible messages and states how many were left out, rather than quietly reading as everything we hold. Deleting a contact or investor still erases every message matched to them, private ones included, and the workspace's retention window purges pulled messages on the same schedule regardless of who could see them.

The reply chase

The follow-up assistant can chase your own unanswered inbound mail: one reminder to you after three days without a reply, a second after seven, then it stops. It reads only mailboxes you connected — never a colleague's — and by default only mail already matched to a contact, company or investor. An opt-in extends it to everything unanswered in your own mailbox; that switch states, before you turn it on, that the scanner reads the subject, sender and dates of unanswered inbound and stores only a reference to the thread and the times involved. It stores no message bodies, and no AI reads your mail unless you click Draft a reply, which runs the same drafting engine the rest of the app uses and never sends anything.

Your chase settings, the per-thread snooze and stop-chasing state, and the log of what each pass looked at and left out travel in the organisation-wide export and are deleted with your account. That history is kept for 180 days and then purged, except where you have told it to stop chasing a conversation — that instruction is kept until you undo it, because an instruction to leave someone alone is not something to forget on a timer.

Logos and photographs

A record can carry a small image beside its name. A company logo is that company's published mark, and it shows by default. A photograph of a person is personal data with no relationship purpose, so it is off unless the workspace turns it on — with it off, no photograph of a person is stored or shown, and contacts and individual investors keep their initials. This app already refuses to collect photographs of people who are not records at all.

Whatever is shown is fetched once, by this app, and served from private storage behind a short-lived signed link. No record image is ever hotlinked from the host it came from, so opening a list never tells that host who was looking, at what, or when. Deleting a record deletes its image with it.

WhatsApp

WhatsApp reaches us three ways, and they hold different data. A business number receives messages people send to it, delivered by Meta as they are sent; it is a company number, so what arrives is visible to the workspace. A chat export is a file somebody uploads from their own phone, and it belongs to them: it follows the same sharing setting as any other connection, and a private one stays readable by its owner alone. A linked device is a member's own personal account, paired the way WhatsApp Web is — the newest of the three, and the one that needs the most said about it.

Chat exports and Business numbers are inbound-only. A linked device may send only a human-composed text reply after a 10-second Undo window; no AI feature can send a WhatsApp message. A linked device additionally never marks anything read, never types, and never joins, leaves or changes anything. The single exception is that unlinking asks WhatsApp to sign the device out too, so it does not linger on the member's account claiming to be connected — the one call this connector makes that changes anything at WhatsApp's end, and the one that removes our own access. It is asked for rather than guaranteed, so the member is also told to check the device is gone in WhatsApp itself. Precisely: like WhatsApp Web it does tell WhatsApp that the device is connected, and acknowledges the messages it receives, because a linked device that does neither stops receiving. It reports the account as offline rather than online, which is what keeps read receipts off. Media is never downloaded — a photo or voice note is stored as a placeholder naming what it was, not as the file itself.

About linked devices specifically. There is no official API for personal WhatsApp, and connecting one this way is not something Meta supports; the account Meta can act against is the member's own. An organisation’s admin can switch it off entirely, and only the member whose number it is can pair it — a session is never shared, handed over or operated by an admin. Only conversations with people already on the workspace's records are imported, and that rule is applied as each message arrives, before anything is written: a conversation with anyone else produces no message and no record that it exists. To be exact rather than flattering, those messages do reach our servers for the moment it takes to discard them; what we promise is that they are never stored. Groups, communities and broadcast lists are never imported at all, because a group is other people's conversation and no one member can consent on their behalf. The session keys that hold the pairing are encrypted at rest and are never shown to anyone, including the member.

The other people in these chats never agreed to be in anyone's workspace, and we treat their data accordingly. Deleting a contact or investor erases their WhatsApp messages with them, exactly as it erases their email, and also clears the remembered link between their name in a chat and their record — their name in that conversation stays, because rewriting who was in a room is falsification rather than erasure. The workspace's retention window purges WhatsApp messages on the same schedule as everything else, and once a chat has no messages left the record of it goes too.

3. Why we are allowed to hold it

For your account with us, we rely on performance of our contract with you, and on legitimate interests for security, fraud prevention and improving the product.

For the business contacts inside your workspace, the lawful basis is yours to establish, not ours — typically legitimate interests in managing business relationships. That means it is your responsibility to be able to justify the records you bring in, to honour objections and erasure requests from the people in them, and to tell those people you hold their data where the law requires it. The product gives you the tools; the obligation is yours.

4. Who else processes it

We use these subprocessors. Several only ever see anything if you switch the relevant feature on.

There is no advertising, no third-party analytics, no tracking pixels of our own, and we never sell or share your data with anyone not on this list.

5. AI, specifically

AI features send the relevant content to Anthropic’s API to produce a result. Only what a given action needs is sent — the record you are enriching, the notes you are summarising — not your whole workspace.

Your content is not used to train models. We use Anthropic’s commercial API, where inputs and outputs are not used for training. Nothing an AI feature produces is written to your data without a person accepting it, and nothing addressed to someone outside your workspace is ever sent without a human reviewing it and clicking send.

6. Where it lives, and how it is protected

On our hosted service your data sits in a Supabase Postgres database in TO CONFIRM — Supabase project region. Files, logos and avatars live in private storage and are served through short-lived signed links, never public URLs.

• Secrets are encrypted at rest. Connector credentials, API tokens and any AI key you store are encrypted with AES-256-GCM before they touch the database, and are never displayed back to anyone, including us.
• Access is scoped by role. Permissions are enforced on the server, not hidden in the interface, and roles can be limited to a single module.
• Changes are auditable. Meaningful changes are logged against the record, with who did it and what the value was before.
• Support access is exceptional and visible. See section 7.

7. When our staff can see your workspace

Day to day, we see account, billing and aggregate usage information — not the contents of your workspace.

To resolve a support issue an administrator can open a time-limited session inside your workspace. Every one of those sessions requires a written reason, expires after 30 minutes, shows a banner inside the workspace the whole time it is active, can be ended by you or revoked by us at any moment, and is permanently logged. Support staff see exactly what the user they are helping can see, and nothing more. Connector credentials and stored keys are never visible, to anyone, in any mode.

8. How long we keep things

Deleting a workspace removes its records in full, including related rows elsewhere in the platform.

9. Your rights

Where the UK GDPR, the EU GDPR or a similar law applies, you can ask for a copy of your data, correct it, delete it, restrict or object to how it is used, and take it elsewhere in a portable format. You can also complain to a regulator — in the UK that is the Information Commissioner’s Office.

Two routes, depending on whose data it is. If you are a customer, exporting or deleting is built in — take the whole workspace as JSON or CSV whenever you like, without asking us. If you are someone whose details sit in a customer’s workspace, that customer controls those records, so ask them directly; tell us and we will help them respond and put you in touch.

Email privacy@minimal.tech and we will respond within one month.

10. International transfers

Some of our subprocessors operate outside the UK and the EEA. Where data is transferred, we rely on the UK International Data Transfer Addendum or the EU Standard Contractual Clauses, together with the safeguards in each provider’s own terms.

11. Cookies

We set a cookie to keep you signed in, and one to remember whether you prefer the light or dark theme. That is the entire list. No advertising cookies, no analytics cookies, nothing that follows you to another site — which is why you are not being asked to dismiss a consent banner.

12. Changes

If we add a subprocessor or change something material about how we handle your data, we will update this page and tell account admins before it takes effect, so you have time to object.

13. Contact

Minimal X Ltd (company number TO CONFIRM), TO CONFIRM — registered office address. Privacy questions: privacy@minimal.tech. Registered with the Information Commissioner’s Office under TO CONFIRM — ICO registration number.