boat

Teams building AI agents or 'agent factories' need isolated, disposable compute environments where agents can run code, browse the web, use Docker, and operate dev tools without affecting production systems. Existing sandbox providers like E2B, Daytona, Modal, Vercel Sandbox, Railway, and others are compared unfavorably on price and performance, costing several times more per month for equivalent VM resources (e.g. $265 to $1,764/mo versus boat's $131/mo for a 100-VM fleet example). Benchmarked build and test loops also run slower on competitor sandboxes (54s to 105s versus boat's 36s on a 4 vCPU/8GB machine). Running this infrastructure yourself requires managing VM provisioning, networking, snapshots, and billing at scale.

boat offers persistent Ubuntu Linux VMs with SSH/SCP access, Docker support, a dedicated IPv4 or IPv6 per machine, a 60fps virtual desktop, and fast disk-level forking from snapshots. Each sandbox comes pre-installed with Docker, VS Code, Chrome, Ghostty, GitHub CLI, Rust, Node.js, and Bun, and can run any agent harness such as Claude Code, Codex, or OpenCode from outside via CLI, API, or SDKs. A CLI (boat new, ssh, scp, exec, prompt, fork, stop, desktop, list) and HTTP API/Python/TypeScript SDKs support both programmatic and human use, with JSONL output via --json for scripting. Billing is per-second with a $20/mo minimum buying 2,000,000 VM-seconds, sandboxes come in three sizes (small, default, large) billed at different rates, stopping a sandbox snapshots it and pauses billing, and plans support 100 to 2,000 active sandboxes with sandboxes runnable in EU regions (Germany, Finland, France).