Agent Node

The Agent node is designed for longer-running, autonomous model work: an agent runtime that can call tools, iterate over multiple steps, and record traces of what it did. Where Text Generation produces a single response from one prompt, an Agent is intended to plan and act across several turns to reach a goal.

What it is for

The Agent node targets tasks that a single generation call cannot complete on its own — work that benefits from tool use, intermediate decisions, and visibility into the steps the model took. The intended model and runtime are configured at the workspace level rather than chosen per-node.

• Tool calling, so the agent can act rather than only respond.
• Multi-step runs that can take longer than a single completion.
• Traces and agent-studio controls for inspecting and tuning what the agent does.

How it differs from Text Generation

• Text Generation is a one-shot call: prompt in, text out, streamed as it generates.
• The Agent is intended to loop — calling tools, reading results, and continuing until it reaches an outcome.
• The Agent surface is oriented around a configured runtime and trace visibility rather than per-node provider and parameter pickers.

Using model-backed nodes today

Until the Agent node ships on the canvas, build with the available generation nodes:

• Text Generation for reasoning, chat, summarization, and code output.
• Prompt to author and reuse the instructions those nodes consume.
• Webhook Call when a step needs to reach an external API.

For the full set of node types and what is available now, see Node types.