🧩 Local-First Agent Sandbox Runtime

This project is positioned as a local-first, self-hosted AI agent runtime and an MCP (Model Context Protocol) bridge. In short, it tries to provide a secure execution foundation for AI agents, featuring sandboxed sessions, memory management, credential storage, and audit/replay capabilities. The hard part is not getting an AI to run, but controlling its execution boundaries within a local environment. It includes a local console to monitor agent behavior. The technical direction is highly relevant right now: as agents become more capable, developers increasingly need a standardized, isolated runtime rather than letting AI operate directly on the host file system. It is more like an exploration of infrastructure standards for the agent era.