Shared AI memory for a small team: what happens when someone leaves
By Matthew Firth, founder of ModelBrain · Oct 24, 2026 · 4 min readA shared AI memory is only as good as its leaver story: what the person who left still has, what the team keeps, and which new work they can no longer see. Here is our own proposal, the leaver test: ask those three questions of any setup before the team puts anything into it.
The leaver test, four setups
| Setup | What the leaver keeps | What the team keeps | What they can no longer see |
|---|---|---|---|
| Repository files (a CLAUDE.md, Cursor project rules) | Any copy already on their computer | Everything committed. Cursor describes project rules as version-controlled and scoped to the codebase | Nothing new once their repository access is removed |
| Cursor user rules | Their own, which Cursor describes as global to their environment | Not stated. The page does not say user rules are shared | Not applicable |
| Claude Team or Enterprise organization | Their account and data. Anthropic says removal does not delete either | Projects shared with the whole organization. Private projects and chats become inaccessible to others | Anthropic's page does not say what happens to a removed member's memory |
| ModelBrain Teams, encrypted shared partitions | What their computers already hold, which a removal cannot take back | The partition, held on every other member's computers and as an encrypted copy on our servers | Cut off on our servers at once. New key at the next sync by someone with write access |
Sources for the first three rows: Anthropic's page on removing a user and Cursor's rules docs. Anthropic's page says custom data retention settings do not cover the memory of members who have left, and says nothing else about it.
What happens in ModelBrain Teams
Per our Teams docs, an owner or admin chooses Remove, and it takes effect at once on our servers. The shared partition then gets a new key the next time someone with write access syncs, so nothing written after that reaches the leaver. One owner or admin confirms the removal in the ModelBrain app, which signs it into the organization's history. Our servers cannot do that on their own. The audit log keeps these events for 400 days and never contains the content of a memory.
Two limits are worth stating plainly. What the leaver already had stays on their computers, as with any shared document, so a removal ends their access to new memories and does not erase old ones. And the docs give no number for the time between removal and the new key, which depends on when a writer next syncs. We have not measured that interval.
A member's personal memories stay theirs. Joining an organization only lets it share partitions with you, and it cannot read your own memories. Shared partitions are end-to-end encrypted with keys held only by members' computers. The shape of that guarantee is the subject of three readers to check before you trust one.
Run the leaver test yourself
- As person A, save a memory to the shared space.
- As person B, in each assistant the team uses, recall it.
- Remove A. As B, save a second memory.
- Check what A can still recall, and from which computer.
What it costs
Teams is billed per node, which is every computer of every active member: $6 a month per node billed yearly or $8 monthly, $4 past 25 nodes, plus $0.05 per GB a month for shared storage. Single sign-on and the audit log are included. If the organization cancels, shared partitions are read-only for 30 days and their encrypted copies are then deleted from our servers, and everyone keeps what is on their computers. Details are on pricing. ModelBrain is in beta and its engine is not open source today.
Sources
- What happens to a user's data when they are removed from a Team or Enterprise organization, Anthropic support
- Rules, Cursor docs
- Teams: share a partition, ModelBrain docs
- Pricing, ModelBrain
Common questions
What happens to shared AI memory when an employee leaves? It depends on where the memory lives. Files in a repository stay with the company, a vendor workspace follows that vendor's removal rules, and a person's private memory may not transfer. Test removal with a second person before you rely on any setup.
Can a removed member still read a ModelBrain shared partition? Their access on our servers ends at once, and the partition gets a new key at the next sync by someone with write access. Anything their computers already held stays there, as with any shared document.
Does removing someone from a Claude Team organization delete their memory? Anthropic's page, checked October 24, 2026, says removal does not delete the account or data. It says custom retention settings do not cover a departed member's memory and does not say more about it.
Last checked October 24, 2026. Settings and tool behavior change often, so confirm a detail on the vendor's own page before relying on it.