CrowNest is cloud sandboxes for coding agents.

Agents write code. That code needs somewhere to run. CrowNest gives it an isolated Linux Sandbox with a real filesystem, real processes, a TTL, and a control plane around it.

Create a Sandbox from a Template. Run Commands and stream stdout/stderr back into your agent loop. Read and write files in /workspace. If the code starts a server, expose it as an authenticated Preview. Promote the outputs you care about to Artifacts, then kill the Sandbox or let it expire.

It fits code interpreters, eval runners, generated app previews, report generation, file processing, test runs, and agentic dev tools where "just run this on our server" is a bad answer.

CrowNest includes the pieces you need to wire it into real workflows: TypeScript SDK, Python SDK, CLI, REST API, MCP server, Vercel AI SDK tools, Workspace Runs for archive-plus-command jobs, Evidence Bundles retained after cleanup, project-scoped API keys, quotas, hard caps, and per-second metering.

Standard runtime is about $0.12 per sandbox-hour, metered by the second. Free starts with a $10 credit, no card.

Use it when your agent needs to install packages, process files, run generated code, or show a private browser preview without getting a long-lived machine or an open-ended cloud bill.

Website: https://crownest.dev

Docs: https://docs.crownest.dev/docs