Skip to main content
Atlas AI agents use language models to answer industrial questions, for example, about historical and planned maintenance or root cause analysis. An agent has instructions, tools, and access to data in the CDF knowledge graph. A project can have up to 100 agents. Each agent can have up to 20 tools. For more information about Atlas AI agents, see About Atlas AI agents.

​
Chat turns

Omit cursor to start a conversation. Pass the cursor from the previous response to continue it. When you send a cursor, do not repeat earlier messages.

​
Tool confirmation

Call Function, Run Python code, and Call REST API wait for confirmation before they run. The response then contains a toolConfirmation action. Send the same cursor and a toolConfirmation message with status set to ALLOW or DENY.

​
Streaming

Set stream to true to receive server-sent events. Each event is an agent session response:
  • progress is a status update while the agent is working.
  • responseChunk is a fragment of the reply. Concatenate the fragments.
  • result is the complete response, including messages, cursor, and actions. It is the last event.
  • cursor is sent at the start when streamOptions.returnCursor is true. Save it so you can resume if the stream stops. Treat result.cursor as the cursor to store.
When stream is false, the response is a single result.
The stable (20230101) API reference does not include AI agents. Use the beta or alpha API reference for /ai/agents/... concepts and endpoint documentation.
Last modified on November 2, 2026