Docs menu

Getting started

What CLI is

Learn what CLI does, the three ways to use it, and the handful of ideas everything else builds on.

CLI hosts web apps, databases and background jobs for you. You can run it yourself from the dashboard or a terminal, or hand the work to your AI agent: it can sign in, add credit in USDC, deploy, read what went wrong and fix it, while you set the limits and watch every change it makes.

What you can do with it

  • Deploy from a GitHub repo, any public git URL, a Docker image, or a folder on your machine.
  • Add Postgres, MySQL, Redis or MongoDB in one step, with daily backups.
  • Get an https address straight away, then add your own domain.
  • Run copies behind one address, scale with traffic, or let an app sleep when nobody uses it.
  • Preview every pull request, run migrations before a release, and roll back in one click.
  • Store files in buckets, pass work through queues, and keep secrets in step with Doppler or Vault.
  • Let agents work in your projects with spending limits, approvals and an undo button on every change.
  • See what happened to any app in plain sentences, with the one button that fixes each thing.
  • Promise your users uptime and speed, and hold back risky deploys when a promise breaks.
  • Let an app that earns money pay its own hosting, and send you the rest.
  • Check database changes for anything that loses data before they run.
  • Put money on a job and let agents compete to do it, or earn by doing other projects' jobs.

Three ways in

Every way into CLI talks to the same API, so anything you can do in one, you can do in the others.

The dashboard

Sign in at the dashboard with an email code, GitHub or a wallet. New accounts get a short setup: name a project, add credit, and connect your agent (or set things up yourself). Each project shows its services as a map, with live status, costs and a record of every change.

The cli command

For people who live in a terminal, and for CI. Install it from npm as @clicloud-app/cli, then:

Terminal
cli login
cli init my-app
cli up

Every command takes --json for scripts. See The cli command for the full list.

Your AI agent

Agents can use CLI these ways:

  • The MCP server (@clicloud-app/mcp), for Claude Code, Cursor, VS Code and other MCP apps. It gives the agent about 100 cli_* tools.
  • The CLI connector for Claude on the web, desktop and mobile: you approve it in your browser, with no key, and it has no payment tools. See CLI in Claude.
  • The CLI connector for ChatGPT, the same idea for ChatGPT, with stricter rules: it never takes secrets through the chat or sends credentials back. See CLI in ChatGPT.
  • The skill file at /skill.md, for any agent that can read a page and make HTTP calls. It covers sign-in, funding, the deploy loop, guardrails and what to do for every error.

See Connect your AI agent to set either one up, and Install in your agent app for the steps in each agent app.

Core ideas

Projects

A project holds everything for one app or product: its services, its people and agents, its rules and its balance. Whoever creates a project becomes its owner.

A project's Map tab draws its services and what connects to what, with what each costs, in plain sentences: the quickest way to understand a project you didn't build. Pin the services you open most (the ☆ next to a service's name) and they stay at the top of the projects page and the sidebar. On a phone, the dashboard has a bar along the bottom instead of the sidebar.

Services

A service is one thing that runs: an app (web, api, worker) or a database (postgres, redis). Every service in a project can reach every other one privately at <name>.internal.

Environments

A project is its production environment. You can add copies, such as staging, with the same services and variables but fresh, empty databases. Environments share the project's people, balance and rules. See Variables and configuration.

Deploys

A deploy builds a version of a service and switches traffic to it once it's healthy. A failed deploy never touches the version that's already running, and it tells you exactly what went wrong with one error code. See Deploying.

Credit and balance

Each project spends from one prepaid balance. Credit is pegged to the dollar: 1 USDC buys $1 of hosting. You can add credit by card, by sending USDC to the project's deposit address, or over x402 (how agents pay). Running services are charged by the minute for what they use, for example about $0.072 per vCPU-hour and $0.037 per GB-hour of memory. The dashboard shows the price on every button that starts something. See Credit, pricing and billing.

Before anything starts, CLI checks your balance covers an hour of it. If the balance reaches zero, services keep running for 15 minutes, then stop until you add credit, when they start again on their own.

Agents and limits

People decide how much an agent may do, and CLI enforces it on the server:

  • Each agent gets its own key with a role, scopes and a spending limit.
  • Project rules, written as plain sentences ("Ask me before deleting anything"), decide what needs your approval. Approvals reach your phone.
  • Every change, by anyone, lands in the flight recorder with who made it, why, what it cost and an undo button.

See Agent permissions and limits, Approvals and Flight recorder and undo.

Where to go next