Skip to main content
Skip to main content

Memory

Beta feature. Learn more.

Memory carries user-specific context across conversations. Rather than indexing entire chat histories, it stores compact structured entries — preferences, recurring facts, project details — that the agent can pull in when relevant.

How it works​

A small memory agent runs alongside the main conversation. It reads recent messages, decides what's worth remembering, and writes entries to a per-user store. On the next conversation, those entries are available as context the main agent can reference without you having to repeat yourself.

You see this as continuity: tell an agent once that you prefer SQL output in lowercase and that your fiscal year ends in March, and future conversations behave accordingly.

Manage your memories​

Open the memory panel from your account menu to:

  • View entries the agent has stored about you.
  • Edit any entry to correct or refine it.
  • Delete entries you don't want carried forward.

Memory is private to your user. Other people's agents never see your entries, and your agents never see theirs.

Toggle memory​

Each conversation has a memory toggle in the chat header. Turn it off for sensitive topics you don't want stored, or for one-off conversations where personalization isn't helpful.

When memory is off, the agent neither reads from nor writes to your memory store for that conversation.

When memory helps​

  • Recurring conventions: preferred date formats, business definitions, naming patterns.
  • Project context: which service or database you usually query, which dashboards you care about.
  • Communication style: terse versus chatty, code-heavy versus prose-heavy responses.

When memory doesn't help​

Memory isn't a database. It's not a place to dump large reference material — use a skill or bake the material into the agent's instructions for that. It's also not retrieval over past chats; the conversation history itself plays that role.