IVY Agent

Build agents that chain tools together to get work done. Describe what you want in plain language and IVY Agent drafts a flow for you, or wire one up yourself on a visual canvas — then run it and watch each step.

What you can do

• Generate an agent from a prompt — describe the task and get a working flow, built from the tools and nodes you actually have.
• Open the flow on a canvas and edit it: add steps, connect them, and see each step's status.
• Add an agent turn where the next step needs judgement, and let a model choose its own tools for that step.
• Run the agent and follow its progress and results in IVY Monitor.

From a conversation

The assistant reads an agent with ivy-agent.describe and changes it with ivy-agent.edit: the file you name, the agent whose editor is in front, or the workspace's one agent. An edit to an open editor is one undo step and is saved; a step the catalogue does not know is refused, and a credential written in the file is never read back.

Requirements

• Generating an agent, and running a turn inside one, needs a configured language model.
• Running one needs the tools and nodes it uses to be available.
• Works in both the browser and the desktop app.

Getting started

Run IvyAgent: Open Canvas to start from an empty canvas, or IvyAgent: Generate from Prompt to have one drafted for you.