Skip to main content
Hive reads the tools your business already uses so it can spot problems and help you act. This page explains what that means for your data: what Hive keeps, who can see it, how long it stays, and the choices you control.
This page describes how the product behaves. Your contract and Hive’s privacy terms govern the legal relationship. For a data processing agreement or a specific question, contact us.

Google user data

Hive’s use and transfer to any other app of information received from Google APIs will adhere to the Google API Services User Data Policy, including the Limited Use requirements. When you connect Gmail, Google Drive, Google Docs, Google Sheets, Google Slides, Google Calendar or Google Analytics, Hive reads that account to answer your questions, raise signals and carry out the actions you ask for or allow in it. Each connected account is set to Read, Draft or Act. On Act, the level new accounts start on, Hive takes reversible actions itself and reports what it did; irreversible or risky actions need your approval on a card. The exceptions, a Signals email reply inside your own thread and the actions you set up in an agent or workflow, are described in Accounts and autonomy. For Gmail, Hive’s first read covers mail from the last 365 days, outside spam and trash.
  • No sale and no advertising. Hive does not sell data it gets from Google, and does not use or transfer it for advertising.
  • Not used to train AI models. Content Hive gets from Google is never included in Help improve Hive, whatever the setting (see The data-sharing setting). Hive’s own decision model for Signals is trained only on invented example mail (see The Signals decision model).
  • Shared only to give you the feature. To answer a request, Hive sends the relevant excerpts to the AI model provider whose model handles it. See Model providers. The connector platform, Nango, holds the account’s access tokens.
  • People at Hive do not read it, except as described under Support access: when you ask us to, or when needed for security or to comply with the law.
  • Session recordings mask it. See Product analytics.
  • You can stop access at any time. Disconnect the account in Hive (see Disconnecting a tool), or remove Hive’s access from your Google Account at myaccount.google.com/permissions.
  • You can have it deleted. Contact support to have the data Hive collected from a Google account deleted. Hive completes the deletion within 30 days of your request.

What Hive stores

Who can see it

  • Only members of your workspace. Every record is tied to one workspace and isolated at the database level. See Security overview.
  • Private items stay private. Chats, files, workflows and agents belong to their creator until shared with a named colleague.
  • A shared chat shares its answers. Sharing a chat shares everything in it, including answers Hive built from your apps, files and Brain. Share only with people who should see that. Everyone the chat is shared with sees every answer in full, in the app and in connected AI tools, and can open the files sent in the chat, read-only. The Library files, apps and agents behind an answer stay closed to them. See Feedback, sharing and history.
  • Feedback goes to the Hive team. A rating or comment you give on an answer, a run, a public agent or Hive itself is sent to the Hive team with your workspace’s name and your role, never your name, your email address or the chat it is about. See Feedback, sharing and history.
  • Email stays with its owner. Mail Hive reads from a connected mailbox is stored as private to the person who connected it. Mail-based signals carry no address, subject or excerpt.
    • The messages a mail signal cites are kept as message ids only, and shown only to the signal’s holder.
    • When you review a reply Hive drafted, the conversation it answers is shown only to you, with email addresses hidden. It is not available over the Hive MCP server. It is no longer excluded from session replay — see Product analytics.
  • Calls stay with their owner. Calls from a connected phone system are visible only to the member who connected it, like their mailbox. See Phone systems.
  • Signals stay with the account they came from. A signal read through a member’s connected account, with its approvals, notifications and receipts, is visible only to whoever holds that account — whichever tool it came from, and whatever the role of the person asking. Owners and Admins cannot see other members’ signals. See Who can see a signal.
  • Every member’s Business Brain is their own. The facts, relationships and learned behaviours Hive builds from a member’s accounts and actions are visible only to that member. Owners and Admins cannot see them, including facts from company systems the member connected. See Who can see what.
  • Personal and sensitive content stays out of the Brain. Before mail, app and meeting events, or what Hive reflects from your work can add people, facts or learned behaviours to your Brain, Hive checks it and keeps out HR matters, health, personal finance, family life, personal legal matters, special-category data (GDPR Articles 9 and 10) and private consumer mail. Rules settle the clear cases and Hive’s Efficient model the rest; when Hive is unsure, nothing is stored, and only a count is logged. On 1,309 synthetic UK and German emails, chat messages, events, tickets, records and facts, averaged over five runs, it kept out 98.3% of sensitive items (96.4% of the items it was never tuned on) and wrongly kept out 9.3% of business items (16.2% of those never tuned on). See What stays out of your Brain.
  • Ask answers stay out of your Brain. Hive does not learn facts from Ask Hive answers: it learns only from actions you approved or declined, and from facts you tell Ask yourself. See Learning from Ask answers.
  • Your first questions are yours. The first-sync progress card and its starter questions are built only from what the member could open in Ask as themselves. Email counts and correspondents come only from the member’s own mailbox. Admins get no wider view: another member’s mail never appears in an admin’s counts or starters. Starter text built from email senders and file names is sanitised, length-limited and never sent to analytics.
  • File permissions travel with files. When Hive syncs Google Drive, Dropbox, OneDrive or SharePoint, it records each file’s access list and respects it when answering.
  • Admins see metadata, not credentials. Owners and Admins can see which connections exist, but not anyone else’s credentials, scopes or actions.
  • Support access. A small number of Hive staff accounts can open a workspace. Hive staff open your workspace only when you ask, and each time is recorded internally. Otherwise they open it only when it is needed to keep Hive secure or to comply with the law.

Retention

Hive does not yet apply one retention period to everything. What is in place today: If your organisation needs a specific retention period, talk to us.

Deleted files

When you delete a chat, remove a file from a chat or a project, delete a Library file or delete a project, Hive permanently deletes each deleted file within about 20 minutes:
  • the file’s copy in storage, including the earlier versions of a Library file;
  • the text Hive read from it, and its search index;
  • for documents created or opened in a chat, their originals, previews and images.
There is no undo or trash. Hive deletes a file only once nothing else in the workspace still uses it. Identical uploads share one stored copy, so a file attached to two chats is kept until both are deleted. A Library file opened as a document in a chat is kept until that chat is deleted too. Files you deleted before this change reached Hive on 3 October 2026 have now been permanently removed in the same way, unless something in the workspace still uses them. Some files are not deleted this way:
  • Files synced from a connected tool belong to that tool. Disconnecting does not delete them; see Disconnecting a tool. When a file is deleted in the tool itself, Hive marks it deleted on the next sync and deletes the search index it built from the file within a few hours, unless the file is still used in Hive.
  • Deleting a Library folder moves its files to the top of the Library; it does not delete them.
  • Deleting a Library template keeps its stored copy.
  • A few stored copies from before 3 October 2026 can no longer be traced to what was deleted, such as pictures inside documents of chats deleted before then, and stay stored. To have them removed, contact support.

Your personal memory

If you choose to use it, Settings → Account → Memory holds short notes Hive remembers about you. They are private to you and used only in your regular chats, and only when your workspace allows chat memory. Delete all memories removes them after you type a confirmation; it does not delete past chats or workspace rules. See Personal memory.

The data-sharing setting

Settings → Workspace → Data sharing has one switch, Help improve Hive. Its wording in the product:
When this is on, Hive will use this workspace’s chat and workflow data to further improve and enhance Hive’s product offerings. No personal data will be used.
  • It is off unless an Owner or Admin turns it on, and can be turned off again at any time.
  • Only Owners and Admins can see or change it.
  • If Hive cannot read the setting, it treats it as off.
  • Content Hive gets from Google services (Gmail, Google Drive, Docs, Sheets, Slides and Calendar) is never included, whatever the setting. This follows Google’s Limited Use requirements. See the Google API Services User Data Policy.
  • If Hive cannot tell where a piece of content came from, it is not included either.
  • Hive tests its own Signals decision model on real workspace data only in workspaces with Help improve Hive turned on, and never on content from Google services.

Product analytics

Hive uses product analytics to understand how the product is used and improve it:
  • Analytics are processed in the EU.
  • IP addresses are not collected, and no location is derived from them.
  • Page addresses in analytics events are reduced to the page’s pattern, for example /ask/:conversationId, so ids and query strings in a link are not recorded as event properties. Page titles and the text of the element you clicked are not recorded either, since a title can carry a chat, deck or file name.
  • Session recordings show layout and clicks, not content. A recording of a signed-in session keeps the page layout, where you clicked and how you moved through Hive, and masks what was on screen:
    • Text is replaced with asterisks, except Hive’s own fixed labels: sidebar navigation, settings sections and page names.
    • Every value you type is masked, and so are labels, tooltips and placeholders.
    • Slide previews, PDF pages and generated finding views are shown as empty boxes.
    • Links in your content are not recorded. Other links and images keep only the Hive page pattern or another site’s address.
    • Request and response headers and bodies are not recorded, and neither is your browser’s console output. A recording keeps only the address and timing of each request, without its query string.
    So your email, files and other content from connected apps, such as Google’s Gmail, Drive, Docs, Sheets and Calendar, do not appear in a recording. Recordings are retained for 30 days and are visible to Hive staff with access to our analytics project.
Recording is tied to your analytics choice. Until you accept on the consent banner, no session is recorded. If you Decline, no session is recorded, and declining is reversible at any time from your preferences.
  • Until you choose on the consent banner, usage is measured anonymously for the current visit only: it is not linked to your account, the session is not recorded, and nothing is stored on your device (no analytics cookie and no stored identifier).
  • Accept lets this browser remember you, links your usage to your account across visits, and starts session recording.
  • Decline turns analytics off in this browser. Hive then records your usage only at workspace level, not linked to you. Browser Do Not Track and Global Privacy Control count as Decline. Usage through Slack, Teams, email and AI clients is always recorded at workspace level.
  • Workspace-level usage is kept under one analytics profile per workspace, which holds no personal data: no member id, name or email address. A personal analytics profile exists only for a member who accepted.
  • The public agent page and public workflow forms report errors only: no banner, no page tracking, no recordings and nothing stored on the visitor’s device.
  • Connection labels such as mailbox addresses are kept out of analytics, and custom integration names are anonymised. For identification and to exclude Hive’s own staff, analytics keep your email address, its domain, your role and your sign-up date — not your name.
  • Hive records when a workspace makes its first connection and gets its first answer that cites a source, as timestamps in the workspace settings. It sends one server-side event for that answer, carrying only the number of sources, the model name and the seconds since the first connection. The onboarding events carry the suite chosen, the number of integrations, the connection count and coarse file and email count ranges (0, 1-99, 100-999, 1,000-9,999, 10,000+). None of them carries a question, an answer, a source, an account, person or file name, a title or file contents.
  • Guided tour. Hive is testing a short guided tour of its screens, filled with pre-made sample data about a fictional client. It is not yet offered to workspaces. Nothing in the tour is sent to an AI model.
    • Hive stores the outcome of your tour (finished, skipped, left part-way, or declined when offered), the last step you reached, the tour’s version and the time. Your browser keeps a copy of the outcome.
    • Each time Hive stores an outcome, it sends one server-side event, module_tour_ended, carrying the outcome, the last step, the tour’s name and version, and whether you had taken the tour before. It is linked to you only if you accepted analytics; otherwise it is recorded at workspace level.
    • The browser sends module_tour_offered, module_tour_started, module_tour_step_viewed, module_tour_action_clicked and module_tour_step_unavailable. They carry the tour’s name, where it was offered or started, the step, its position and the number of steps, which sample control you used, why a step could not be shown, and your language and text direction.
    • None of these events carries an id, a person’s or company’s name, or anything you typed.
    • While the tour runs, the browser sends no other Hive usage events, and the events the analytics tool records on its own carry module_tour_running: true, so they can be told apart from real use.
  • Signals pipeline health. Hive sends one server-side event for each hourly signal check, each reminder sweep and each slice of reading a member’s connected accounts for matters. Each carries the workspace and only counts, yes/no flags and coarse ranges, for example items read, matters created or updated, signals raised, how late a reminder ran, and a rough range of AI use and cost. It is grouped by workspace, never by person, and never carries a member, account, mailbox, matter, title, signal or message text. These events never carry raw token counts or spend.
  • AI model calls. Hive also records metadata about each model call: the model, the part of Hive that made it, token counts, timing, cost and a fixed error code. It is kept at workspace level, and linked to you only for a request you made in the web app after you accepted analytics. It never carries a prompt, a completion or anything read from your data. See What Hive records about each model call.
  • Operational logs. Hive’s operational logs for signal detection and hand-offs carry ids, counts, detector names and error classes, never a correspondent, subject, record name or error text read from a member’s mailbox, files or connected accounts.

Disconnecting a tool

When you disconnect an account (Integrations → the tool → Disconnect):
  • Hive stops reading from it immediately.
  • Open signals that account was feeding are dismissed and their pending approvals removed.
  • For a file store or mailbox, Hive deletes its sync position for the account: how far it had read the files or mail, and the list of items it still had to retry.
  • Nothing you already approved is undone.
  • For MCP servers and incoming webhooks, the stored secrets and pinned tools are deleted.
Disconnecting stops future reads. It does not delete the data Hive already collected from the account, so that it comes back if you reconnect the same account. If you want that data removed, contact support: Hive completes the deletion within 30 days of your request. For a Google account, you can also remove Hive’s access from your Google Account at myaccount.google.com/permissions. Hive can then no longer read the account.

Erasing someone from the Business Brain

When a person asks to be forgotten, Hive can permanently erase them, or any other entity such as a customer or supplier, from your Business Brain. Erasure deletes:
  • the entity itself;
  • every fact about it, including superseded history, and its learned baselines;
  • the alternative names Hive knows it by;
  • every relationship that points to it by its id, name or an alternative name, current or past, including relationships recorded on other entities;
  • the links from that member’s matters to the person, whether the link names the entity, their email address or their name.
Erasure also removes the person’s name from the text Hive stored about them in that member’s matters and matter Signals, and puts a removed person in its place (a removed company for a customer, supplier or other company, and a removed record for anything else). That covers matter titles, summaries, positions and next steps, the matter timeline, the labels of linked items, requirements and hand-offs, and the title, explanation, stake line and evidence of each Signal raised on those matters and its Inbox card. Their email addresses are removed from the matter’s matching keys and from requirements, so new mail from them is no longer filed on the matter. To avoid rewriting text about someone else, Hive matches the person only by their full name, by an alternative name of two or more words, or by an email address it held for them. It never matches a first name on its own: erasing “Jane Doe” leaves “Jane”, “Mary-Jane Doe” and “Jane Doe-Smith” as they are. A person known to the Brain by a single word and no email address has no text rewritten. Erasure cannot be undone. Every member has their own Brain, so an erase reaches one member’s Brain at a time; to have someone erased from every member’s Brain, say so when you ask. It needs the Owner or Admin role, and your own business’s entity cannot be erased. There is no erase action in the app yet: to have someone erased, an Owner or Admin can contact support.
Erasure does not rewrite other facts whose text merely mentions the person, Signals that were not raised on one of that member’s matters, or another member’s matters. It does not touch mail or documents stored in your Library, chats, or the connected tool the data came from. If the source record still exists in a connected tool, Hive can learn it again on a later run, so remove or correct it there too.

Leaving Hive

Hive support deletes a workspace: an Owner or Admin can contact support, as Settings → Workspace → General → Delete workspace says. When a workspace is deleted, Hive:
  1. engages the workspace’s kill switch so nothing can act;
  2. removes the comment attachments members had already deleted, and makes a last pass to delete the files they had deleted (see Deleted files);
  3. stops all scheduled jobs for the workspace;
  4. cancels its subscription at once and ends its plan. Usage the workspace already owes is invoiced, the time left in the current period is not refunded, and its unused credits are voided. Credit top-ups still waiting for payment are cancelled; a top-up whose payment goes through after the deletion adds no credits and is refunded in full.
  5. stops live updates from its connected apps: Hive ends the change notifications it set up with Outlook and Google (mail, calendar and drive), and ignores events other apps still send. The connections themselves are kept, not deleted, so a restored workspace keeps its connected accounts and Hive turns their live updates back on automatically.
To keep using the workspace until the end of the period you have paid for, cancel the subscription in Settings → Billing → Plan with Cancel subscription instead, and delete the workspace after it ends (see Trial and payments). Once 30 days have passed, a daily clean-up permanently deletes the workspace’s Business Brain: its entities, every fact including superseded history, baselines, relationships and alternative names. Restoring the workspace within those 30 days cancels this and keeps the Brain. The workspace’s other records, including billing records, are retained so it can be restored if you return. To request full deletion or an export of your data, contact support. Hive completes a deletion request within 30 days.

Security overview

Isolation, encryption and policy.

AI and models

How AI providers handle your data.

Infrastructure

Where Hive runs.

Health and disconnect

Connection status and removal.