Skip to main content
Long-term memory lets the agent keep facts that stay true beyond one conversation, such as a target’s technology stack or what has already been tested, and read them back in later chats. It is off by default. Open it from Settings → Memory in the AI Studio sidebar. The header reads “Long-term memory content used by the AI agent” and holds the On/Off switch.
Agent settings Memory section with an Off switch in the header and a notice reading Long-term memory is off. The agent does not save memories across conversations. Turn it on to let the agent remember important information.

Settings → Memory while long-term memory is off

Whose memory it is

Memory is per member, per workspace. The agent only reads and writes your own memories in the current workspace; other members cannot see them, and your memories in one workspace are not used in another. Only the main agent uses long-term memory. Subagents never read or write it; within a conversation they share findings through short-term memory and the workflow notebook instead (see Side panel).

Turn memory on

1

Open Memory

In AI Studio, open Settings → Memory.
2

Switch it on

Click the switch in the header so it reads On. The console confirms with “Long-term memory turned on”.
3

Keep chatting

From your next message on, the agent can save memories and sees the ones already saved. Saved memories appear in this section.
To stop, click the switch again (“Long-term memory turned off”). Changing the switch requires the agent.write permission.

What the switch does

What gets remembered

There is no form to add memories: the agent decides what to save while it works. It is told to keep knowledge that is true about the workspace beyond the current conversation, one fact or topic per memory, for example:
  • the technology stack, authentication model or WAF behavior of a target
  • what has already been tested, and what was found
  • known vulnerabilities and organizational policies that affect how to test
You can also ask for it directly, for example “Remember that the staging environment is out of scope”. To correct or remove a memory, ask the agent: it can list memories with their IDs, replace one with updated content, or delete one. Each turn, the agent’s context includes your 50 most recently updated memories. When you have more, it is told how many are hidden and can list the rest on demand.
Memories are stored as workspace data, not as secrets. Do not ask the agent to remember passwords, API keys or tokens.

Review and delete memories

With memory on, the section lists your memories, newest first. Each entry shows its content (long entries are cut to three lines with Show more / Show less) and when it was last updated. The list shows your 10 most recently saved memories; ask the agent to list them all.
1

Find the memory

Make sure memory is On, then find the entry in Settings → Memory.
2

Delete it

Click the trash icon on the entry, then confirm Delete in the Delete memory dialog. Deleting cannot be undone.
Memories cannot be edited from settings. To change one, ask the agent to update it, or delete it and let the agent save a new one. See Permissions.

Instruction

Standing instructions you write yourself

Chat

Start a conversation with the agent