AI Agent Glossary

Plain-English definitions of every term you'll see in the OpenClaw guide.

AI agent: An AI agent is software that can perceive its environment, make decisions, and take actions autonomously to achieve a goal. Unlike a plain chatbot that only responds to messages, an agent can use tools (calendars, browsers, file systems), run multi-step plans, and operate without human input at every step. OpenClaw is an AI agent framework.
LLM (Large Language Model): A Large Language Model (LLM) is a neural network trained on huge amounts of text. Examples include GPT-4, Claude 3.5, Llama 3, and Mistral. OpenClaw uses an LLM as its reasoning engine — you choose which one to plug in.
Local model: A local model runs on your computer instead of a vendor's server. Local models are slower than cloud models but fully private — no data ever leaves your machine. Common local models include Llama, Mistral, and Qwen, served via tools like Ollama or llama.cpp.
Cloud model: A cloud model is hosted by a third party (OpenAI, Anthropic, Google) and accessed through an API. Cloud models are typically faster and smarter than local models, but each message you send leaves your machine and is processed by the vendor.
Skill: A skill is a small module that gives OpenClaw a new capability — for example, reading your calendar, searching the web, controlling smart-home devices, or editing files. Skills are how you customize the agent for your specific needs.
MCP (Model Context Protocol): Model Context Protocol (MCP) is an open standard for connecting LLMs to external tools and data sources. OpenClaw supports MCP servers, which means thousands of off-the-shelf integrations work out of the box.
Sandbox mode: Sandbox mode is a restricted environment in which OpenClaw can run code or shell commands without being able to touch the rest of your system. The guide covers how to enable and audit the sandbox.
Self-hosted: Self-hosted means you run the software on your own machine or your own server, instead of paying a vendor to host it for you. Self-hosting trades convenience for privacy, control, and (usually) lower long-term cost.
Hermes: Hermes is another open-source AI agent framework. It's powerful but more developer-oriented than OpenClaw, with steeper setup and a smaller skill ecosystem. The guide includes a side-by-side chapter comparing the two.
VPS (Virtual Private Server): A VPS is a small virtual computer you rent from a cloud provider (DigitalOcean, Hetzner, Linode) for $5–10/month. It lets OpenClaw answer your messages 24/7 without leaving your home machine on. The guide includes a step-by-step VPS setup chapter.
OpenClaw: OpenClaw is a free, open-source AI agent framework that runs on your own computer or a small VPS. It connects LLMs (cloud or local) to messaging apps like WhatsApp, Telegram, and Discord, plus a growing library of skills for productivity and automation.
WhatsApp bridge: A WhatsApp bridge is the software layer that lets OpenClaw read and send WhatsApp messages on your behalf. The guide walks through installing and pairing the bridge in Chapter 3.