OpenClaw Glossary
- Agent
- A software entity — usually powered by a large language model (LLM) — that perceives its environment, makes decisions, and takes actions to accomplish goals. OpenClaw agents can control your operating system, browse the web, write code, and more.
- Agentic AI
- A category of artificial intelligence where the AI system acts autonomously, planning and executing multi-step tasks with minimal human intervention. OpenClaw is an example of agentic AI.
- ClawHub
- The official community repository for OpenClaw skills. Think of it like a package registry (npm, PyPI) but for agent capabilities. Browse, install, and publish skills from ClawHub.
- CLI (Command-Line Interface)
- A text-based interface for interacting with software. OpenClaw provides a CLI (
claw) for running agents, installing skills, and managing configuration from your terminal. - Context Window
- The maximum amount of text (measured in tokens) that a language model can process in a single request. Larger context windows let agents handle more complex instructions and longer documents.
- Guardrails
- Safety mechanisms that prevent an agent from taking harmful or unintended actions. OpenClaw supports configurable guardrails like file-system restrictions, network allowlists, and confirmation prompts.
- Hallucination
- When an AI model generates information that sounds plausible but is factually incorrect. OpenClaw reduces hallucination risk through tool-use verification and skill-based grounding.
- LLM (Large Language Model)
- A neural network trained on massive text datasets that can generate, summarize, translate, and reason about text. GPT-4, Claude, Gemini, and Llama are examples. OpenClaw can connect to multiple LLMs.
- MCP (Model Context Protocol)
- A protocol that manages how context (instructions, memory, tool results) is structured and passed to the underlying language model. MCP ensures the agent always has the most relevant information in its context window.
- Multi-Agent
- A system architecture where two or more AI agents collaborate, delegate, or compete to complete tasks. Related platforms like BitBuddies focus specifically on multi-agent orchestration.
- Plugin
- A software extension that adds new functionality. In the OpenClaw ecosystem the term "skill" is preferred, but "plugin" is sometimes used interchangeably (especially in older docs).
- Sandbox
- An isolated execution environment that prevents an agent from modifying the broader system. OpenClaw can run tasks inside sandboxed containers to limit blast radius if something goes wrong.
- Skill
- A modular, installable capability for an OpenClaw agent. Skills are published on ClawHub and can include anything from "browse the web" to "deploy a Docker container." Install with
claw install <skill-name>. - Token
- The smallest unit of text processed by an LLM — roughly ¾ of a word in English. Token counts determine cost and context window limits. "OpenClaw" is typically 2–3 tokens.
- Tool Use (Function Calling)
- The ability of an AI model to invoke external tools (APIs, code execution, file I/O) rather than relying solely on text generation. OpenClaw agents heavily leverage tool use for reliable task execution.
- Workflow
- A defined sequence or graph of tasks that an agent executes. OpenClaw supports workflow chaining, where the output of one skill feeds into the next. See Usage → Workflow Chaining.
Missing a Term?
This glossary covers the most common terms. For deeper technical details, check the OpenClaw GitHub docs or visit the ELI5 explainer for simplified explanations.