daytonaio/daytona ? reverse-engineered prompt

Build me an open source platform like Daytona where AI agents can safely run generated code inside isolated cloud sandboxes. I want users and agents to be able to create a sandbox, run Python, JavaScript, or TypeScript code, read and write files, install tools, stream logs, and stop or delete the sandbox when done.

Make it feel like a real developer product, with a simple API, a command line tool, and a web dashboard where someone can see active sandboxes, open a terminal, inspect files, view running processes, and manage snapshots so work can continue later. Sandboxes should be Docker compatible, isolated from each other, and configurable with CPU, memory, disk, environment variables, and base images.

Please build a working MVP with clean setup instructions, local development support, basic auth or API keys, clear error handling, and example scripts showing an AI agent creating a sandbox and running code. Look up current docs online if you need to.