Skip to content

Data model

XTM One is easier to understand when you treat it as a set of connected objects rather than a set of disconnected screens. The same object can appear in Chat, the Agentic Flow graph, History, and the AI Catalog without becoming four different things.

At a high level there are three kinds of things: assistants that do work, resources those assistants rely on, and records that show what happened.

Assistants and records

  • Agents are the main assistants. An agent bundles its instructions and persona, tools, knowledge bases, prompts and skills, assignments, handovers, and memory. If you remember one object type, remember this one.
  • Assignments are automation rules inside an agent — they define what triggers the work and what the agent does when it runs. They are not separate assistants.
  • Runs are the execution records created when assignments or other background work happen. They tell you whether the work ran, is still running, failed, is waiting for a person, or what it produced — the operational layer of the product.
  • Conversations are the chat sessions between a user and the General Assistant or a specific agent. The conversation holds the discussion; the agent stays the reusable assistant behind it.

Reusable resources

The Resources area holds the building blocks agents and chats depend on: knowledge bases (searchable documents), prompts (reusable instructions), skills (reusable instruction packages), MCP servers (external tools over the Model Context Protocol), tools (built-in or custom actions), and integrations (connections to outside systems).

Access and sharing

Users have personal settings and access; groups are used for controlled sharing. Most objects have one visibility state — personal, shared with a group, or company-managed — which decides who can edit them and whether you should reuse an object or copy it.

How it appears in the UI

Most areas follow the same pattern: a list page, a detail page, and tabs for overview, configuration, activity, and related records. The UI is not inventing a new concept on every screen — it shows different views of the same underlying objects. In practice they chain together: a user opens an agent, the agent contains assignments and references resources, running it creates a run in History, and chatting with it is stored in a conversation.

Next step

Continue with Relationships and nesting, which explains how these objects point to one another and where configuration actually lives.