Skip to main content
Extend the default Atlas agent for your project’s terms and skills, and build a separate agent only when a workflow needs its own instructions, tools, or language model.

About planning an Atlas AI agent

Decide whether a workflow fits an agent, and shape its scope and instructions.

About language models

Match a language model to the agent’s task complexity, speed, and cost.

About skills

Give the agent domain knowledge that loads only when the question needs it.

About tools

Choose a focused set of tools, and route calculations to tools instead of the language model.

About subagents

Let one agent delegate a task to another independently configured agent.

Build and publish agents

Create and publish an agent in the Agent builder.

Configuring skills

Write, structure, and test skills with the skill builder or by hand.

Build agents with the Cognite CLI

Create and push agent files, run evaluations, and deploy through version control and CI/CD.
Last modified on November 17, 2026