Memory
Memory modes and scopes, how memory is read before the answer and written after it, and the approval policy.
Memory makes an agent remember across messages and conversations.
Modes (Memory tab)
- Off: nothing is remembered beyond the conversation history.
- Remember in this chat: items are scoped to the conversation.
- Remember for this agent: items are shared by everyone talking to this agent.
- Remember this user: items are scoped to the user (or end user) across their conversations.
Scopes available in workflow nodes: user, conversation, agent, app and workspace. A namespace separates memory sets within a scope.
Reading memory
Before the model answers, the memory read node searches for relevant items (recent, keyword, hybrid, importance, vector or semantic hybrid strategies) and injects them into the prompt under "Relevant memory:" with their source and score.
Writing memory
After the answer, the memory write node extracts candidate items (preferences, facts, project context, instructions) from the user input and the assistant output. The default write policy is candidates require approval: candidates appear in the memory items list where a team member approves or rejects them. Automatic saving above a confidence threshold can be enabled per agent. Sensitive content is excluded by default, duplicates are merged, and items can update existing ones or expire after a retention period.