Agent memory

Some agents are set up with a long-term memory store — a knowledge context the agent can read from and write to on your behalf. When such an agent learns something worth keeping, it saves a memory item so the fact is available in future conversations too. Not every agent has memory enabled. If yours does, you’ll see it ask a short question before it saves anything.

​

The save card

When the agent finds something worth keeping it shows a Remember this? card instead of asking in prose. The card contains the wording (editable), the memory type, and who may see it — Private (only you) by default, or Public (everyone using this agent), with the same access control you know from knowledge items. Nothing is saved until you choose Save. Don’t save tells the agent to drop it. If the agent proposes several memories at once you can Save all or decide one by one.

​

Corrections and forgetting

Tell the agent “that’s wrong, it’s …” or “forget that”. It proposes an update or a deletion on a card you confirm. You can only change memories you saved yourself or have write access to; for other people’s public memories the agent tells you who saved it so you can ask them.

​

Seeing what was used

Answers that used memories show Recalled N memories under the reply, with each memory, who saved it and when. From there you can open, edit or forget your own memories. The header chip Remembers N things about you opens the panel with everything the agent remembers that you saved, split into Private and Public. Builders and curators review everything an agent has learned under Build → Memory: every memory base, its memories, and each memory in full. See Memory area.

​

What the agent can remember

The shape of a memory item is defined by the knowledge context the agent is connected to. Every memory includes a name, a description, and a surrounding context note that records why the fact was saved (for example, which question or topic triggered it). Depending on how the agent is configured, extra fields — such as category tags, a rating, or a project label — may also be filled in. These fields are set by whoever built the agent; you do not need to worry about them. What matters from your side is the visibility choice above.

​

Where memories live

All memory items are written to the knowledge context the agent is configured to use as its memory store. You can find them in the Knowledge workspace:

1. Open Knowledge from the main navigation.
2. Locate the context that the agent uses as its memory store (the agent’s name usually gives a hint).
3. Browse, search, edit, or delete any item there.

To delete a memory, open the item and choose Delete. The agent will no longer retrieve that fact.

​

For builders

The long-term memory setting for an agent — including which knowledge context serves as its memory store — is configured in the agent workbench. See Agent workbench for details.