Team Memory

Privacy in plain words

Team Memory remembers what your team decides in group chats. Here is exactly what that means for your messages.

What we read

Only the chats an admin of your team connected: groups, and private chats they picked one by one. Nothing else on a linked WhatsApp or Telegram account is read.
Team Memory announces itself in every chat it reads: a notice in groups, a hello in private chats. There is no secret listening.
Anyone can reply /hide to a message in Telegram or WhatsApp to keep it out. Admins can /pause note-taking at any time.

Raw messages, for the number of days your team chose (90 by default). Then they are deleted automatically.
What the AI found in them: decisions, ideas, promises, open questions and topics, with links to the original messages while those still exist.
Reminders that were planned and sent, and who pressed Done, Snooze or Not important.

An owner can switch this on in Settings → Privacy. Chat messages are then deleted about an hour after the AI has read them, and so are messages hidden with /hide.
Copies of chat text, such as the reply assistant’s log and reminders that were sent, are deleted after a day, so there is still time to review or delete a reply.
Decisions, promises, ideas and questions are kept.
It’s off by default. With it on, the reply assistant sees less of each conversation, and learns your writing style from fewer of your messages.

Every team’s messages are encrypted with the team’s own key. The keys are kept in a separate vault, locked with a key that isn’t stored with the data, so a copy of the database on its own shows only scrambled text.
Only people in your team can open your team’s messages. Other teams, and anyone outside your team, get nothing. Team Memory’s AI opens them only to do its work: finding what was decided and promised, and writing the replies you allowed.
Linked WhatsApp and Telegram logins and Claude keys are stored encrypted too, and everything travels over encrypted connections.
What stays readable, so search and the dashboard work: the notes found in your messages (decisions, promises, ideas, questions), and who wrote each message and when.

Only people in your team who have a Team Memory login. Every team’s data is separated inside the database itself.
Team Memory is about topics, not people: there are no “who talks most” charts or productivity scores.
Personal promises are reminded privately, never in front of the group.

Messages are read by Anthropic’s Claude to find decisions, ideas, promises and questions: with your team’s own Claude API key, or through a team member’s own Claude subscription.
We never use your chats to train AI models. With a Claude subscription, messages go through that person’s own Claude account, so their Claude privacy settings apply.
The AI is told to skip personal and sensitive matters, and every result links back to its source so people can check it.

Owners can start over in Settings → Privacy: delete only the chat messages, or everything learned from them, while keeping the team and its connected chats.
Owners can also delete everything at once, the team included. It happens immediately and can’t be undone.
Disconnecting a chat in Settings → Chats stops it being read straight away.

Supabase (database and background jobs), Anthropic (AI), Railway (the always-on WhatsApp and Telegram connections), Resend (email, if enabled) and Vercel (website hosting).

The full privacy policy and terms are at /legal/privacy.