Skip to main content

Quickstart

Get an AI agent running in an isolated sandbox in 3 steps.
1

Install the CLI

Verify the installation:
2

Authenticate

Sign up at caged.dev, then log in:
Alternatively, set the CAGED_API_KEY environment variable — useful in CI/CD.
3

Create a Sandbox

Option A: Quick start with flags
Option B: Config-as-code (recommended for teams)Create a .caged.yaml in your project root:
.caged.yaml
Then just run:
The CLI reads .caged.yaml automatically — no flags needed.

What Happens Next

Sandboxes boot in ~300ms using Firecracker microVMs. Each gets its own Linux kernel — full OS-level isolation.
  1. A Firecracker microVM boots with your chosen template
  2. Your repo is cloned into /workspace
  3. The init_script runs (e.g., npm install)
  4. The sandbox is ready for your agent

Connect to Your Sandbox

Monitor in the Dashboard

Open caged.dev/dashboard to see:
  • Live terminal — watch what the agent is doing
  • File browser — see files as they change
  • Cost tracker — real-time spend in dollars
  • Trust score — behavioral safety rating
  • Preview URLs — click to see web apps the agent builds

Clean Up

Next Steps

Concepts

Understand sandboxes, sessions, and trust scores

Config-as-Code

Full .caged.yaml reference

Agent Setup

Connect Claude Code, Cursor, and other agents

CLI Reference

All CLI commands