Glossary

📚 Glossary: Definitions of OpenClaw terms, concepts, and technical jargon. Use this page when reading docs or community discussions.

Overview

This glossary defines key terms used in the OpenClaw ecosystem: project names, architecture components, security concepts, and related AI and messaging vocabulary. Terms are listed alphabetically. For step-by-step guides, see the Quick Start and Installation hub.

A

Agent
An AI system that can take action beyond replying with text. In OpenClaw, agents run 24/7, execute tools (browser, files, shell, email, calendar), and maintain memory. Contrast with a chatbot, which mainly answers questions. See AI Agent vs Chatbot.
Anthropic
Company behind the Claude family of LLMs. OpenClaw supports Claude as a model provider via API. The project was briefly named “Clawd” as a play on Claude + claw.
API key
A secret token used to authenticate with an LLM or service (e.g. Anthropic, OpenAI). In OpenClaw, API keys are configured per model provider and should be stored in environment variables or a secrets manager, not in config files. See Security Best Practices.
Autonomous (agent)
Capable of acting without a human in the loop for each step. OpenClaw agents can run cron jobs, proactive tasks, and multi-step workflows without constant prompting.

B

Bind address
The network address the gateway listens on. Best practice is to bind to localhost (127.0.0.1) only and access remotely via SSH tunnel or VPN (e.g. Tailscale) to avoid exposing the gateway to the internet.
Channel
A connection to a messaging platform (WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Teams, etc.). OpenClaw uses channel adapters to send and receive messages. You configure one or more channels in config. See Channel Setup.
Channel adapter
Pluggable connector that talks to a specific messaging platform. Each adapter handles that platform’s API, auth, and message format so the gateway can route messages to the same agent across platforms.

C

Chatbot
An AI that primarily converses in text. Unlike an agent, it typically does not execute tasks (browse, run commands, send email). OpenClaw is an agent platform, not just a chatbot. See AI Agent vs Chatbot.
Clawd
Original name for the project (November 2025), a play on “Claude” and “claw.” Renamed to Moltbot after an Anthropic request, then to OpenClaw (January 2026).
ClawHub
Community marketplace for OpenClaw skills. Users can browse, install, and publish skills. Security research (e.g. Snyk) has found some skills with credential leaks or malicious behavior; audit skills before use. See Skills Marketplace Guide and Security Best Practices.
Clawdbot
Earlier name for the project; sometimes still used in search. Now OpenClaw.
CLI (command-line interface)
The openclaw command-line tool used to install, configure, start/stop the gateway, run openclaw doctor, and manage channels and config. Requires Node.js 22+.
Cron (job)
A scheduled task that runs at fixed times (e.g. daily morning briefing, hourly checks). OpenClaw supports cron-style proactive tasks so the agent can act without a user message.

D

Data sovereignty
Keeping data under your control and jurisdiction. Self-hosting OpenClaw supports data sovereignty because conversations and logs stay on your infrastructure instead of a third-party cloud.
Daemon
A background service that runs continuously. The OpenClaw gateway runs as a daemon, managing connections, routing messages, and keeping the agent alive.
DeepSeek
LLM provider supported by OpenClaw as a model provider. Often used for cost-effective or regional deployments.
Docker
Containerization platform. OpenClaw can be deployed via Docker for consistent, isolated environments. Best practice: avoid mounting docker.sock; use minimal volumes and resource limits. See Docker Deployment and Security Best Practices.

E

Extension
Another word for skill in the OpenClaw ecosystem. Skills extend the agent with new capabilities (e.g. web search, code execution, custom APIs).

G

Gateway
The core OpenClaw service (daemon) that accepts connections from channels, routes messages to the agent, runs the LLM and skills, and manages sessions and memory. It should run on your own machine or VPS and typically bind to localhost only.
Gemini
Google’s LLM family. OpenClaw can use Gemini as a model provider.

L

LLM (Large Language Model)
The AI model that powers reasoning and text generation. OpenClaw is model-agnostic: you can use Anthropic Claude, OpenAI GPT, Google Gemini, DeepSeek, or local models (e.g. via Ollama).
LM Studio
Local LLM inference tool. OpenClaw can be configured to use models run through LM Studio for fully local, private inference.

M

Memory (system)
OpenClaw’s persistent context storage. The agent can remember information across conversations and channels, enabling long-term context and personalized responses. Configured in Configuration / Memory.
Model-agnostic
Not tied to a single AI model. OpenClaw works with any compatible LLM (Anthropic, OpenAI, Google, DeepSeek, local), so you can switch or combine providers.
Model provider
The service that supplies the LLM (e.g. Anthropic, OpenAI, Google, DeepSeek). You configure one or more providers and set API keys. See Model Provider Setup.
Moltbot
Name used for the project (late December 2025) after the Clawd rebrand. “Molt” refers to molting (growth). Rebranded to OpenClaw in January 2026.

N

Node.js
JavaScript runtime. OpenClaw requires Node.js 22 or newer. The CLI and gateway are built on Node.js.

O

Ollama
Tool for running LLMs locally (e.g. on Mac or Linux). OpenClaw can use Ollama as a model provider for fully local, private AI with no API cost.
OpenClaw
Free, open-source autonomous AI agent platform. It runs on your own infrastructure (Mac, Linux, Windows, Raspberry Pi, VPS), connects to messaging apps (WhatsApp, Telegram, Discord, Slack, Signal, iMessage, Teams, and more), and uses an LLM to act-not just chat. Formerly Clawd and Moltbot. Created by Peter Steinberger; governance transitioning to an open-source foundation. See What is OpenClaw?
OpenAI
Company behind GPT and ChatGPT. OpenClaw can use OpenAI models as a model provider via API. Peter Steinberger announced in February 2026 that he would join OpenAI; the OpenClaw project continues under community governance.

P

Prompt injection
An attack where malicious or crafted input tries to change the model’s behavior (e.g. to leak system prompts or perform unintended actions). Defenses include using stronger models, input validation, and clear boundaries for untrusted content. See Security Best Practices.
Proactive task
A task the agent runs without a user message, e.g. scheduled via cron or triggers. Enables 24/7 automation like morning briefings or monitoring.

R

RPC (Remote Procedure Call)
How the CLI and other clients talk to the gateway. If the gateway is unreachable or tokens mismatch, you may see RPC or probe errors; use openclaw doctor and check Troubleshooting.

S

Sandbox (mode)
A restricted execution environment that limits what the agent can do (e.g. block certain shell commands or file access). Enabling sandbox and tool allowlists/denylists is a recommended security practice.
Self-hosted
Running software on your own hardware or VPS instead of a vendor’s cloud. OpenClaw is self-hosted: your data and agent run on machines you control, improving data sovereignty and privacy.
Session
A conversation or interaction context. OpenClaw manages sessions per user/channel and can persist context via memory. Session path or schema issues can cause config validation errors.
Skill
An extension that adds capabilities to the OpenClaw agent (e.g. web search, code execution, custom APIs). Skills are installed from ClawHub or built locally. Audit skills for security before use. See Skills Marketplace and Creating Custom Skills.
Peter Steinberger
Austrian developer who created OpenClaw (originally Clawd/Moltbot). In February 2026 he announced joining OpenAI; the OpenClaw project is transitioning to open-source foundation governance.

T

Tool
An action the agent can perform (e.g. run shell command, read file, send email, query API). Skills register tools. In sandbox mode, you can allowlist or denylist tools for safety.

V

VPS (Virtual Private Server)
A remote server you can use to run OpenClaw 24/7. Common providers include AWS, DigitalOcean, Linode, Hetzner, Vultr. See VPS Deployment and VPS Comparison.

W

WSL2 (Windows Subsystem for Linux)
Recommended way to run OpenClaw on Windows. You run the CLI and gateway inside a Linux environment on Windows. See Windows Installation.

Key Resources

Community & Official Links