run-llama/sandboxed-lit

sandboxed-lit is a small command-line tool written in Rust that lets you point an AI agent at a folder of files and ask it questions about them, while keeping the agent confined to an isolated virtual machine so it cannot affect the rest of your system. The isolation comes from microsandbox, a lightweight VM technology. When you run sandboxed-lit, it creates or reattaches to a contained environment with limited CPU and memory, and mounts your chosen directory into that environment at a fixed path. The AI agent running inside can list files, read them, and execute shell commands, but only within that sandbox boundary. What makes it useful for document work is the file reader. Plain text files are returned as-is, but PDFs, images, and Office documents are routed through LiteParse, a separate tool from the same organization that extracts structured content from those formats. This means the agent can read a PDF or a Word document and reason about its contents without you having to convert it first. The agent itself uses OpenAI GPT models via a Rust SDK called agent-sdk. You give it a prompt on the command line and it runs until it has completed the task, streaming its progress to the terminal with colored output. Someone would use this to have an AI summarize documents in a folder, extract information from a batch of PDFs, or run analysis scripts in a safe environment. The tech stack is Rust with an OpenAI API key required.