Memory

Access remembers what it learned from your chats.

That isn’t a file you uploaded. After a conversation, Access can keep the useful facts — how you like a summary, which table to trust, what “the billing cluster” means — and bring them into the next one. You can read every note, correct it, or dismiss it.

Two places it lives

You. Open Settings → Memory. Things remembered about you from your chats. They’re added to future conversations unless a chat is marked temporary.

A named agent. Open the agent → Memory. Operational facts that job has learned. They’re added to future conversations with that agent.

The default Access agent uses your memory. A named agent uses its own.

How it works

1. Talk in Chat as usual.
2. On a schedule — or when you press Dream now — Access looks through the chats it’s allowed to use.
3. It keeps what it’s sure about. Uncertain notes are skipped. You can still write a correction yourself.
4. The next chat with that scope starts with those notes already in mind.

💡 Tip: A chat’s menu can Exclude from memory or Make temporary. Temporary chats don’t train memory and don’t receive it.

What you can change

• Turn dreaming on or off per scope
• Set the timezone and the time it looks through chats
• Correct a note in your own words
• Dismiss a note so it isn’t injected again

Clearing memory deletes remembered notes from chats and turns dreaming off. Source chats and anything you wrote stay.

Memory isn’t files

Files are the library you give agents — runbooks, skills, working files. Memory is what Access kept from conversation. Use both: upload the policy; let the agent remember that you always want the short version.