Agent Framework
Sign Up Now →Sign Up Now

Build an agent. Put it to work.

Create agents from instructions, skills, and tools. Deploy them together as an application, then follow their work, decisions, and requests for human approval.

Available

Your first agent: Start with a small job you can check.

A support assistant or refund helper gives you a concrete starting point. Shape its instructions, try it with sample data, and inspect what it would do before deploying.

Start the walkthrough →

Choose your starting point#

The pieces you work with#

Agent
A named assistant with a model, instructions, skills, and a declared set of tools.
Instructions
The agent’s job, voice, and boundaries. These guide its behavior; enforced rules control governed actions.
Skill
A reusable procedure or reference in a SKILL.md file that the agent can consult.
Tool
A declared operation the agent can call, such as reading an order or issuing a refund.
Application
A bundle of one or more agents, deployed and versioned together. Its application.yaml records membership and permitted messaging.
Draft and live
A draft is the version you are editing. The live application changes only when you deploy.

Which version does this guide cover?#

This guide follows the application-based CLI and console as of 12 October 2026. It covers Build, Applications, Agents, Runs, Inbox, and Usage. Your administrator supplies the console address and access to the supported agent CLI.

The local CLI service supports a smaller feature set than the product server. Work lists require the product server. This guide is not a hosted-service sign-up or production installation guide.