Command line

Agent CLI — build AgentUI apps from your terminal

The Agent CLI pulls your AgentUI projects onto your laptop as plain files, so you edit them in VS Code or Cursor, keep them in git, and push changes back. Every command also takes --json, which is what makes it drivable by Claude Code, Codex, or Aider rather than only by a person.

Read the docs on npm View on GitHub

MIT licensed. Node 22+. Nothing to sign up for to read the docs.

Terminal

@agentuiai/cli

$ npm install -g @agentuiai/cli
$ agentui auth login
$ agentui project create --name "Ops Dashboard"
  → ./ops-dashboard  (AGENTS.md + skeleton)

$ agentui --json project list
{
  "total": 2,
  "projects": [
    { "name": "Ops Dashboard", "type": "APP_V2" },
    { "name": "Nightly Report", "type": "AUTOMATION" }
  ]
}

Every command speaks JSON, so an AI agent can drive it

Pass --json on the root command and every subcommand returns machine-parseable output on stdout, errors included — the same approach the AWS and GitHub CLIs take. That is the difference between a tool an agent can use and one it has to screen-scrape.

sync, edit, push — your project as files you own

sync writes your pages, components, functions, and entity schemas to disk as .jsx and .schema.json. push compares the working tree against a local manifest and sends only the difference — and --dry-run shows you exactly what it would send before it sends anything.

$ agentui project sync <workflowId>
  components/   pages/   functions/   entities/

$ code .                          # edit with anything
$ agentui project push --dry-run  # see the diff first
$ agentui project push

What it will not do

Secret values never leave the server — not in secrets list, not in a sync bundle. Path traversal is rejected on every read, write, and delete. Every endpoint checks that your company owns the workflow being touched, and a cross-tenant request returns 404 rather than 403, so it does not even confirm the resource exists.

Command reference

agentui auth	Log in with an emailed 6-digit code; check or clear the session.
agentui workspace	List the workspaces you belong to and select the active one.
agentui project	create, list, sync, init, status, push — the core loop.
agentui env	Read and switch the environment a project is pointed at.
agentui secrets	List and request secrets. Values stay server-side.
agentui validate	Check a project for errors before you push it.
agentui build	Build the project bundle.
agentui deploy	Ship a built project.
agentui skills	Manage the skills available to a project.
agentui packages	Manage npm packages a project depends on.
agentui blueprints	Work with prebuilt app blueprints.
agentui prompts	Manage the prompts a project uses.
agentui functions	Inspect and manage server functions.
agentui integrations	List and configure third-party integrations.
agentui report	Report something broken or missing, from the terminal.

Full reference, flags, exit codes and troubleshooting live in the README on npm. @agentuiai/cli

Reading this as an AI assistant? The whole command surface is one flat markdown file: /agentui-cli.md

Install it and sync your first project

npm install -g @agentuiai/cli

Read the docs on npm

MIT licensed. Node 22+. Nothing to sign up for to read the docs.

What an agent can do with these commands — and what it cannot

Using ChatGPT or Claude in the browser instead? See AgentUI for AI agents