Get started

Install the CLI, run an agent on your machine without a model key, then build a real one, either by asking your coding agent or command by command.

You need Python 3.12 or 3.13 and uv. Everything up to deployment runs locally; a Kubernetes cluster is needed only to deploy.

• Installation & setup

  Install the CLI from PyPI, add the skills to your coding agents and check your environment with login.

  Install graph-agents-cli

• Quickstart

  Five minutes: create a project, ask it a question and run the eval gate on the deterministic fake model, then switch to a real provider.

  Run your first agent

• Pick a tutorial

  Build an agent that calls an API with a policy and an approval, evaluate it and deploy it to a local cluster, with your coding agent or by hand.

  Build with a coding agent · Manual workflow

Understand the lifecycle

Every project follows the same path: create, develop, evaluate, deploy, operate. The lifecycle explains each stage, the commands that belong to it and what their exit codes mean.

• The lifecycle

  The stages, the commands in each, and how a change moves from your laptop to production.

• Guides

  Task-by-task depth once the basics work: auth, the API policy, approvals, deployment.

• CLI reference

  Every command and flag, generated from the CLI itself.