Memory Overview

Memory lets the CubePlex agent remember information across conversations. Instead of repeating yourself every time you start a new chat, the agent recalls your preferences, project facts, and team conventions automatically.

Three tiers of memory

CubePlex organizes memory into three scopes. Each scope controls who can see and benefit from a memory item.

Tier	Who sees it	Best for
Personal	Only you, in any workspace within your org	Individual preferences, corrections, personal notes
Workspace	All members of the workspace	Project facts, team procedures, codebase conventions
Organization	Everyone across all workspaces	Company-wide policies, brand guidelines, shared decisions

When the agent processes your conversation, it pulls in relevant items from all three tiers. The memory block presented to the agent groups items by scope (organization, then workspace, then personal) and labels workspace and organization items as user-contributed, so the agent can weigh them appropriately. Within any single scope, correction items take priority over ordinary memory of the same topic.

Memory types

Each memory item is classified by type, which helps the agent understand how to apply it:

• preference — How you or your team likes things done. "I prefer TypeScript over JavaScript."
• project_fact — A concrete fact about a project, codebase, or domain. "Our API uses snake_case for all JSON keys."
• procedure — A step-by-step process to follow. "To deploy, run make build then make deploy-staging before pushing to production."
• correction — Something the agent got wrong that you corrected. "PostgreSQL, not MySQL — we migrated last year."
• decision — An agreed-upon decision that should be respected going forward. "We chose Tailwind over CSS modules for the new dashboard."
• org_policy — An organization-wide rule or standard. "All public APIs must include rate limiting."

How memory works

The agent handles memory automatically in two directions:

Recall — At the start of each conversation turn, the agent retrieves memory items relevant to the current context. You do not need to ask it to "check memory" — it does this on its own.

Storage — When you share important information during a conversation, the agent may save it as a memory item. This happens when:

• You explicitly ask: "Remember that our staging URL is staging.example.com."
• You correct the agent: "No, we use Poetry, not pip."
• The agent identifies a reusable fact during conversation.

Each stored memory item includes:

• Content — The actual information.
• Scope — Personal, workspace, or organization.
• Type — One of the six types listed above.
• Source — A reference back to the conversation, run, or artifact that created it.
• Confidence — A score from 0.0 to 1.0, self-rated by the agent, indicating how confident it is in the memory's accuracy. New items default to 0.8.
• Status — Active (used in recall) or archived (hidden from recall but not deleted).

Next steps

• Using Memory — Learn how to teach the agent and shape what it remembers.
• Managing Memory — Review and archive memory items from the Memory Center.