OpenClaw Glossary

Agent (AI Agent)
An autonomous AI system capable of reasoning, using tools, and accomplishing multi-step goals without continuous human prompting (e.g., your OpenClaw setup).
Anthropic
An AI research company that created the Claude family of models, heavily favored for coding, nuanced writing, and large context windows.
API (Application Programming Interface)
The bridge that allows two software systems to communicate. OpenClaw uses APIs to talk to OpenAI, Anthropic, or external tools like Telegram and Stripe.
ChatGPT
The popular consumer-facing web interface created by OpenAI, powered by their GPT base models. Distinct from the OpenAI API, which developers use to build tools like OpenClaw.
Context Window
The maximum amount of text (measured in tokens) an AI can "remember" and process in a single interaction. For example, a 128K context window is roughly a 300-page book.
Gateway
The structural core server of OpenClaw. It routes messages securely between your LLM, your installed skills, and your user interfaces (like Slack or Telegram).
Human-In-The-Loop (HITL)
A security paradigm where an AI is paused before executing a destructive or high-risk action (like making a payment or deleting files) until a human clicks "Approve."
LLM (Large Language Model)
The "brain" of modern AI. A sophisticated neural network trained on massive amounts of data to understand, generate, and reason with human language.
LM Studio
A popular desktop application that allows you to discover, download, and run LocalLLMs directly on your Mac, Windows, or Linux machine through a user-friendly interface.
LocalLLM
An LLM that is hosted entirely on your own local computer hardware rather than in the cloud. Running a LocalLLM ensures 100% privacy, zero subscription costs, and offline capabilities.
MD File (Markdown)
A lightweight text file format ending in .md. It uses simple syntax (like # for headers and * for bold) to format text natively. OpenClaw uses .md files heavily to store agent memories and instructions.
Model
A specific version or "flavor" of an AI (e.g., GPT-4o, Claude 3.5 Sonnet, Llama-3). Different models are optimized for different tasks—some for speed, some for math, some for creativity.
Ollama
A lightweight, developer-focused command-line tool that lets you run, pull, and manage open-source models (LocalLLMs) directly from your terminal.
OpenAI
The AI research lab behind ChatGPT and the GPT series of models (GPT-3.5, GPT-4, GPT-4o). Their models are widely integrated into cloud-hosted OpenClaw agents.
OpenClaw
A powerful, customizable, open-architecture AI agent framework intended to give individuals and businesses absolute control over their autonomous systems.
Prompt Injection
A cyber attack where a user maliciously formats their text to override an AI agent's original safety guidelines. Security skills like Prompt Guardian defend against this.
Skill
A modular piece of code or capability given to an OpenClaw agent. For example, a "Calendar Skill" gives the agent the ability to read and create Google Calendar events autonomously.
Soul (SOUL.md)
In OpenClaw, the SOUL file acts as the absolute defining system prompt. It dictates exactly who the agent is, its personality, its core beliefs, and its overarching objectives.
Token
The basic unit of data an LLM reads and generates. A token is roughly equivalent to a syllable or word fragment. APIs bill users based on token usage.
Vector Database (RAG)
A specialized database that converts human text into numbers (vectors), allowing an AI to instantly search thousands of documents to find relevant information before answering a question.
Workspace
The secure directory folder where your OpenClaw agent lives. It contains the agent's memory, tools, configuration files, and any outputs it generates over time.
ClawUniverse
Explore the full ClawUniverse ecosystem.
Explore ClawUniverse →