ironclaw

Why we included this project

Most AI assistants either run inside a chat app or hand your data to a cloud vendor. IronClaw takes the opposite route: you install it on your own machine (it supports macOS, Linux, and Windows/WSL), bring your own LLM API key, and it keeps credentials, memory, and tool state in an encrypted local store. You interact with it through a browser UI, a REPL, or channels like Telegram and Slack, and you can schedule background routines on cron or event triggers so it handles monitoring and maintenance on its own. The part that stands out is security: third-party tools run in a WebAssembly sandbox with capability-based permissions, endpoint allowlisting, and secret leak detection, so the assistant can act without every tool getting full access to your system. It also extends through the MCP protocol and dynamically built WASM tools, which makes it a solid choice for developers who want a local agent host they can audit and build on.