AI memory
The AI remembers durable facts from tasks and chats so they don't have to be explained again in every conversation – things like "primary buttons always use the brand colour", "we don't deploy on Fridays" or "customer X is internally called project Y".
You can view, extend and export this under Settings → Personal → AI settings → Memory.
Two scopes
| Scope | Visibility |
|---|---|
| My memory | Only for you – personal preferences and ways of working. |
| Team memory | Team-wide for everyone in the organization – conventions, decisions and stable references. |
Writing to team memory by hand (scope Team) requires admin/owner. The AI also adds team entries from other members' conversations, though – following the rules below. Whoever contributed such an entry finds it under "My contributions to the team memory" and can edit or delete it there.
What gets stored
Every entry has a kind so the AI knows how to treat it:
| Kind | Example |
|---|---|
| Preference | "Answer concisely, without a preamble." |
| Convention | "Commit titles start with the item number." |
| Decision | "We stay on Postgres, no Mongo." |
| Reference | "Design files live in the Figma project 'Liza UI'." |
| Context | "I work in the Europe/Berlin time zone." |
Preferences and context belong in your own memory. The AI only puts conventions, decisions and references into the team memory.
Each entry also shows where it came from (chat, task, conversation or manual), how often it has been used, and – if set – when it expires.
What may go into the team memory
Every AI in the organization reads the team memory – so it reaches everyone who works with the AI. That is why the AI only stores unproblematic things there: conventions, decisions and stable references such as colours, terminology, the tech stack or processes.
Never in the team memory:
- anything personal about individual people – performance, health, salary, HR matters, conflicts, private life
- confidential details about customers, contracts or finances
- credentials and other secrets
- content from meetings, calls or transcripts – at most an unproblematic decision taken there
If the AI is not sure a fact is fine for the whole team, it asks first or saves it for you only.
Private conversations
A switch in the header of the AI chat shows whether the conversation contributes to the memory (Public) or not (Private). From a private conversation the AI saves nothing – neither for you nor for the team.
Adding something yourself
Via "+ Add fact" you enter a rule directly instead of waiting for the AI to learn it. You choose the kind and the scope (Only me or Team). Entries can be deleted again when they no longer hold.
Export
The memory can be exported as JSON or Markdown – useful for archiving or for handing it to another tool.
Retention
How far back the memory reaches depends on the plan (on Free the last 50 days, unlimited on paid plans) – see Plans & Limits.
For developers
Reachable programmatically via the MCP tools save_memory, search_memory and get_org_overview – see AI memory (MCP).