Skip to content
Water LLM

Security & your data

Water LLM reads your Gmail, Calendar, and Contacts. That is a lot of trust to ask for, so here is precisely what is kept in your Google account, what the AI features send to Google Cloud, what reaches our servers, and how long we keep it.

Kept in your Google account, never on our servers

  • Your contacts and every CRM row
  • Your emails
  • Your calendar events and meeting notes
  • Your notes, tags, and custom fields
  • Invoices, clients, and deal records
  • Your LinkedIn export and message history

Reaches our servers

  • Your email address, as the subscription key
  • Your billing status and plan details
  • Stripe customer and product identifiers
  • The date you accepted the licence terms
  • The ID of your Water LLM spreadsheet
What the AI features send to Google Cloud. To write briefings, summaries and drafts, some features send part of the data above to Google Cloud’s Vertex AI — only what each feature needs: calendar titles and attendees, CRM fields, and for the features that summarise correspondence, excerpts of the emails themselves. Prompts and AI responses are never used to train, fine-tune or improve any AI model — a contractual commitment covering all Vertex AI customers. All processing happens inside Google Cloud; no third-party AI provider is involved, and Water LLM retains no copy. Google may briefly retain request data for abuse monitoring under its standard Vertex AI terms; that data is not used for training and Saville Coaching cannot access it. The breakdown, feature by feature, is in Data & Privacy.
Looking for the permission list? Every Google permission the add-on asks for, what each is used for and why a narrower one would not work is on Permissions. This page is about what we store; that one is about what the add-on can reach.
Your business, data and relationships run inside Google. The add-on is Google Apps Script executing under your credentials, your CRM is a spreadsheet in your own Google Drive, and the AI features run on Google Cloud. There is no copy of your relationship data on our infrastructure to leak, subpoena, or lose — because we never store it.

Every field we store

Our database holds one record per subscriber, keyed by email address. This is all of it.

WhatWhy it exists
Email addressThe key that ties your subscription to your Google account, so the add-on knows you have access.
Subscription statusActive or inactive. This is the only field the add-on actually reads.
Billing stateActive, past due, cancelling, cancelled, or expired — so support can tell you what happened to your account.
Plan detailsMonthly or annual, the amount, and the currency.
Stripe identifiersCustomer, price, and product IDs, so your payments reconcile with your access.
Billing period datesWhen your current period started and when it renews or ends.
Cancellation timestampsWhen a cancellation was requested and when access actually ends.
Payment failure timestampsSo we can give you a grace window to update your card rather than cutting you off immediately.
Licence acceptanceThe date and version of the terms you accepted.
Spreadsheet IDThe identifier of your Water LLM sheet. An identifier only — it grants us no access to the contents.
Signup dateAnchors the onboarding email sequence.

Alongside that we keep a queue of the onboarding emails we have sent or are scheduled to send you, which contains your email address and nothing else.

How it is protected

No client can read it

The database denies every read and write from browsers and client applications without exception. Only our own server processes can reach it, and they authenticate as trusted service identities.

Hosted on Google Cloud

The subscription database runs on Google Cloud, encrypted at rest and in transit, in the same infrastructure that holds your Workspace data.

AI stays inside Google

All AI processing goes through Google’s Vertex AI — the same infrastructure behind Google Workspace AI. No OpenAI, no Anthropic, and no other third-party AI provider ever sees your data.

Logs carry no customer content

Our diagnostic logs record which job ran, whether it passed, how many records it touched and how long it took — not what was in them. Names, email addresses and calendar titles are masked before anything is written. There is a full example below.

What our logs actually look like

Every claim on this page is easier to check against an example than a description. Below is what one hour of Water LLM’s background automation produces — the thing we actually read when something has gone wrong for someone. It is built from the exact log lines in our source code, showing a deliberately bad hour so that the error cases are visible too. It is not a capture from any real person’s account, because publishing one of those would be the very thing this page says we do not do.

That is the whole vocabulary: a timestamp, which job ran, whether it passed, how many records it touched, how long it took, and the error if there was one. Where a record has to be identified — row 47, contact index 412 — it is a position in your spreadsheet, which means nothing to us because we cannot open your spreadsheet. Where a value would otherwise appear, you get its length and nothing else: <redacted:14> was somebody’s name.

Why a length and not a scrambled version of the name. A scrambled name can be unscrambled by whoever holds the log, because there are only so many names to try. A length cannot. It tells us the field was not empty and roughly how big it was — which is what actually helps us find a bug — and tells us nothing about who it was.

These logs live in Google Cloud, are tagged with an anonymous identifier Google generates rather than your email address, and are kept for 30 days. One honest caveat: when a Google service returns an error, its message occasionally quotes the value that caused it back to us. We do not control the wording of those messages. It is rare, it is bounded to error text, and we would rather write it here than claim a guarantee we cannot enforce.

Retention and deletion

EventWhat happens
You cancel Access continues until the end of the period you have paid for. The record is then marked for deletion and removed automatically about 28 days later — a window that exists so a cancellation made in error, or a change of mind, does not cost you your setup. Entries in the onboarding email queue expire within 30 days.
You ask us to delete your data You can do this yourself at any time, from Configure Hub → Danger Zone → Delete My Data inside the add-on — it removes the record immediately and cascades to the onboarding email queue. If you cannot reach the add-on, email us and it is done within 30 days. Either way, it does not cancel billing; cancel that separately.
You uninstall the add-on Water LLM immediately loses all access to your Gmail, Calendar, and Contacts. Your spreadsheet stays yours, fully intact and usable.
If you stop paying, you keep your data. Your CRM is a normal Google Sheet in your own Drive. Cancel and it carries on working as a spreadsheet, with every contact, note, and invoice still in it. There is nothing to export, because nothing was ever taken away.

Read the full privacy notice Ask us a question