Agent memory

Agents can keep short notes you tell them: a tone you prefer, a convention your team uses, a fact about a table. Later chats start with the notes that matter for them, so you don't repeat yourself.

What is stored

A memory is a short key and value, at most 1 KB, sometimes tied to a table, an integration, a workflow or a skill. Agents are told never to save secrets, instructions found in data, or anything about one person. Appenda refuses text that looks like a key or token. It also refuses a note about a table that names a person or a single record in it.

Where it lives

• Local workspaces: in the workspace's encrypted file on your computer.
• Cloud workspaces: in Appenda Cloud, in your workspace's region.

Who can see each memory

Admins can delete another member's memories without reading them.

How saving works

Agents save automatically, using appenda.memory_save, and show it in the chat. A save made after the agent read anything else in that turn (a table, a web page, an integration's results) waits for you to keep it. Until you do, it is listed but never recalled. Saves in a bot's routine runs always wait.

Each chat turn starts with your pinned notes, the workspace's required rules and the notes about the table you're looking at. Agents find the rest with appenda.memory_search. Every note reaches the agent as quoted data, never as an instruction.

Delete or turn off

• Memories can be listed, kept, pinned and deleted per workspace. The Settings → Agents → Memory page that shows them is coming in a later release.
• Agent memory can be turned off for a whole workspace by someone who manages its agents. While it is off, no notes reach a chat and the memory tools refuse. Turning it off deletes nothing: turning it back on brings every note back. The Settings switch for this is coming with the Memory page.