Memory

Memory is notes an agent keeps between conversations. It is stored as a tree of nodes, each with a title and some content, inside a memory category. Open Knowledge sources → Memory to see and edit it yourself.

Memory differs from documents: you and the agent write it as short notes, and it is organised as a tree rather than as files. To turn memory on for an agent, see Memory and chat history.

Memory categories#

The page opens on the list of memory categories. Memory categories are separate from the document categories on the Categories page.

• Click Create Category, give it a Category Name and a Category Color, and save.
• Click a category to open its tree.
• With a category open, Edit Category renames it or changes its color. The same dialog has Remove Category.
• Refresh reloads the data, for example after an agent has written new nodes.

Work with nodes#

A new category is empty. Click Add Parent node to create the first node. Hover over any existing node and click the plus button (Add child node) to create a node under it.

In the form, enter a Node Title and Node Content, then save.

Click a node in the tree to open it on the right. You see its Path and Content. From there you can:

• Click Edit Markdown or Edit plain text, change the content, then Save.
• Delete the node, after a confirmation.

Use the search box above the tree to find a node.

How agents use memory#

In an AI agent, turn on Enable Memory and choose a Memory Category. The agent refuses to run with memory on and no category chosen. Then pick the Access Mode:

• Read Only: the agent can search memory.
• Read/Write: the agent can also add and update nodes.

The agent also has read and write prompts, which tell it when to use its memory tools. You can edit both. Two agents that share a memory category share what they remember.

Next steps#

Read Memory and chat history for the agent side of memory, or Categories for document and Q&A groups.