Docs

Persistent memory.

The agent remembers across sessions, including preferences, site quirks, and past outcomes, so it gets faster and more accurate the more you use it. All of it is inspectable and yours to edit or clear.

How it works

Instead of starting cold on every task, the agent writes durable notes to a persistent store and recalls the most relevant ones when a new task begins. That's why the second time you ask it to work in a familiar app, it already knows the layout and your preferences. Every entry is human-readable, with no black box.

What's stored

  • Your task preferences and how you like results formatted
  • Layouts and quirks of sites you visit often
  • Outcomes and summaries of prior tasks, for context on the next one
  • Named contexts you create to group work by project

Inspecting and editing memory

Open Settings, then Memory to browse everything the agent has stored. From there you can:

  • Search entries by keyword, site, or task
  • Edit an entry to correct a preference or an out-of-date detail
  • Delete individual entries you no longer want kept
  • Pin a note so the agent always weights it highly

Clearing memory

You can remove a single entry, clear one named context, or wipe all memory from Settings, then Memory.

Clearing memory is permanent and can't be undone. Wiping everything resets the agent to a first-run state, so it won't recall any past preferences or sites.

Scopes

Memory is scoped per account and per workspace. Personal preferences stay with your account. Team workspaces share only the task context relevant to that workspace, keeping personal data separate. Named contexts let you narrow recall further, useful when one project shouldn't bleed into another.

Privacy

Memory exists to serve you, and you control all of it: everything is visible, editable, and deletable at any time. Sensitive actions still pass through approval gates regardless of what the agent remembers. A stored preference never bypasses your approval for a purchase, login, or message.

Related