Native agents are agents you create yourself inside Dart. Give your agent a Name and an Icon so your team can recognize it, then choose how it should function. There are four types:

Internal agents that run on guidelines you write describing the agent's personality and what it does. Best for "do anything in Dart" workflows like parsing task descriptions, estimating sizes, breaking work into subtasks, triaging bugs, or organizing your workspace. You also pick the model that powers it (Auto/Dart AI, Claude, ChatGPT, or Gemini).

Connect an external agent you access through an API. Define the endpoint to send data to and the trigger that starts the agent (such as commenting at it). This is the path for setting up your own integration with an outside service.

Run a coding tool on your own computer, such as Claude Code, Codex, Gemini, or OpenCode, instead of through a cloud service. Useful when you want to keep the coding session close to your local repo, credentials, and environment.

An agent identity with no built-in automation, driven entirely through the Dart API. Best when you want full control of the agent's behavior from your own code.

Native agents fit into your normal Dart workflow. Assign one to a task, mention it in a comment, or chat with it. As soon as it's the assignee, the agent reads the task, follows its configuration, and executes the work, updating statuses based on your settings.