nearai/ironclaw

What it solves

IronClaw is a secure, personal AI assistant designed to give users full control over their data and AI interactions. It addresses the lack of transparency and data privacy in corporate AI systems by ensuring all information is stored locally and encrypted, preventing data harvesting and telemetry.

How it works

IronClaw operates as a local service that connects to an LLM provider of the user's choice. It uses an "Agent Loop" to handle messages and route intents. To execute tasks, it employs a highly secure environment: untrusted tools run in isolated WebAssembly (WASM) containers with capability-based permissions and endpoint allowlisting. For more complex jobs, it uses a Docker sandbox. Memory is managed via a workspace filesystem with hybrid search (full-text and vector search) for persistent context.

Who it’s for

It is designed for individuals and professionals who want a private, auditable AI assistant that can be customized with self-expanding capabilities and integrated into various channels like Telegram, Slack, or a web browser.

Highlights

• Defense in Depth: Implements multiple security layers, including prompt injection defense, credential protection, and leak detection to prevent secret exfiltration.
• Self-Expanding: Users can describe new tools, and IronClaw builds them as WASM tools on the fly, or connect to Model Context Protocol (MCP) servers.
• Multi-channel Access: Supports interaction via REPL, HTTP webhooks, WASM channels (Telegram, Slack), and a dedicated Web Gateway.
• Background Automation: Features a Routines Engine for cron schedules, event triggers, and webhook handlers for proactive maintenance and monitoring.