Skip to main content
Memory lets the agent store and recall information across conversations. Tell it your brand colors once and every website, deck, and carousel it builds starts from those defaults. Tell it your company name, tech stack, or tone of voice and it remembers. The agent checks its memory at the start of every task. You do not need to repeat yourself.

How memory works

The agent uses a memory system in its sandbox. It can store text and images, search through stored memories, and delete ones that are no longer relevant.
1

You tell the agent something worth remembering

Set brand preferences.

2

The agent stores it in memory

The agent runs a memory command in its sandbox to save the information. It is linked to your account and persists across all future conversations.
3

Future conversations start from your preferences

Before building anything, the agent checks memory for relevant context. Your brand colors, company name, tone, and preferences are applied automatically.

What you can store

TypeDetails
TextUp to 5,000 characters per memory. Brand guidelines, preferences, project notes, tech stack, anything.
ImagesStore images with optional captions. Useful for logos, reference designs, or style guides.

When the agent checks memory

The agent checks memory before every task. It looks for relevant preferences based on what you asked. If you ask it to build a website, it checks for brand colors, design preferences, and tech stack. If you ask for a report, it checks for formatting preferences and company details.

What you cannot do

  • You cannot organize memories into categories or folders. All memories are in a flat list.
  • You cannot share memories with other users.
  • You cannot set expiration dates on memories. They persist until you delete them.
  • You cannot store files other than images (text and images only).
  • You cannot store memories longer than 5,000 characters each.
  • You cannot bulk-delete memories.

Next steps