gitmyhub

ironclaw

Rust ★ 12k updated 15h ago

IronClaw is an Agent OS focused on privacy, security and extensibility

IronClaw is a privacy-first personal AI assistant that runs entirely on your own machine, keeps your data encrypted locally, and isolates every plugin in a secure WebAssembly sandbox so tools cannot leak your credentials.

RustWebAssemblyPostgreSQLsetup: hardcomplexity 4/5

IronClaw is a personal AI assistant you install and run on your own machine. The core idea is that your data stays on your device, encrypted, and never gets sent to a company's servers without your knowledge. The project is open source and built in Rust, a programming language known for reliability and speed.

The security design is one of the more distinctive parts of this project. Any tool or plugin that IronClaw runs is executed inside a WebAssembly sandbox, which is an isolated container that limits what the tool can actually do. Credentials like passwords or API keys are never handed directly to tools; they are injected at a controlled boundary so tools cannot read or leak them. The system also checks outgoing web requests against an approved list of destinations, blocking anything that was not explicitly permitted.

You can talk to IronClaw through several different interfaces: a command-line prompt, a browser-based web interface with real-time streaming, or messaging platforms like Telegram and Slack (connected via plugin channels). It can run scheduled tasks on a timed schedule, respond to incoming webhooks from other services, and handle multiple jobs at the same time in isolated contexts.

One feature the README highlights is the ability to build new tools on the fly. You describe what you need, and IronClaw generates a new WASM tool to do it without requiring a software update from the developers. It also supports the Model Context Protocol, which is a standard for connecting AI assistants to external data sources and tools.

For the AI brain itself, IronClaw defaults to NEAR AI but lets you swap in other providers including Anthropic, OpenAI, Google Gemini, Mistral, and locally-run models via Ollama. Configuration is handled through a setup wizard that walks you through connecting a PostgreSQL database (required for memory and search), authenticating with your chosen AI provider, and storing secrets in your system keychain.

Installation is available via a shell script for macOS and Linux, a Windows installer, or Homebrew. Building from source requires Rust and a database setup step. The license is MIT or Apache 2.0.

Where it fits