Hey Product Hunt, I’m Tristan! I’m building CrowNest. CrowNest gives coding agents real Linux sandboxes. The loop is pretty straightforward: create a sandbox, run code, install packages, stream stdout, work with files, open a private browser preview when the code starts a server, save the outputs you care about, then shut the machine down. I built it because agents keep increasingly need somewhere to do real-world work. They need to run generated code, install dependencies, write files, start servers, and bring back useful results. I did not want that to mean handing every agent a permanent devbox, a local machine, or a blank cheque for cloud compute. The launch version has TypeScript and Python SDKs, a CLI, REST API, MCP server, and Vercel AI SDK tools. Also scoped API keys, projects, usage visibility, spend caps, authenticated previews, and durable artefacts. I’m especially curious what people building coding agents, code interpreter flows, eval systems, or agentic dev tools would need before moving a real workflow onto this. Also, what should come next: richer templates, persistence/resume, better terminal controls, or easier browser handoff (or something else)? Docs: https://docs.crownest.dev

CrowNest

Real Linux sandboxes for coding agents

Real Linux sandboxes for coding agents