Agent memory

Each agent has an agent memory where the AI notes procedures and caveats learned from conversations with that agent. Manage it in the Memory section of the agent’s configuration panel.

​

How it works

• Once a chat with this agent has settled, the AI reads it back and writes down what is worth remembering. During the chat itself it writes only when you ask it to remember something
• What it writes down is shared with everyone who uses this agent
• A chat that settles after you turn learning on can be read including the exchanges from before you turned it on
• In chats started without an agent, the AI can also consult the memories of the agents you can use (read-only). Up to 8 memories are consulted in one chat, including your personal memory; beyond that, recently updated agents take priority

​

Turning learning on and off

Use the “Learn from chats (shared with everyone who uses or edits this agent)” switch in the Memory section.

• Turning it off stops new entries from being written (existing memories remain and are still recalled in chats)
• You can also ask from chat — “turn off this agent’s memory” — and the switch flips only when you approve the confirmation card

​

Viewing and managing memories

The Memory section lists everything this agent has learned.

• Members with edit permission can edit or delete each entry individually, or delete every entry at once with “Delete all” (a confirmation is shown; this cannot be undone)
• View-only members (“Use only” permission) can read the entries but not change them

Erasing memory — a single entry or all of it — happens only on this screen. Asking from chat will direct you here instead.