OpenClaw is a free, open-source AI agent that bridges the gap between LLMs knowing what to do and actually doing it. It runs as a local Node.js service using a hub-and-spoke architecture centered on a gateway that handles message routing, session management, and tool use. The agent operates on the ReAct pattern: reasoning over context, acting via tools (browser, terminal, APIs, skills), and observing results in a loop until a task is complete. Users interact through messaging platforms like Slack, iMessage, or Discord via adapters. Skills are markdown-based instruction files that extend the agent's capabilities on demand without flooding the context window. Security risks include misconfigured internet-exposed instances, prompt injection from untrusted inputs, and unsafe skill code — requiring isolated environments, credential encryption, and careful code review.
Sort: