A
AES-256-GCM Encryption
Military-grade encryption used to protect stored data. It means that even if an attacker gained access to your server's hard drive, your data would be unreadable without your encryption key. Used by IronClaw for all locally stored data.
API (Application Programming Interface)
A way for software to communicate with other software behind the scenes. When Open Claw "calls" Claude or GPT to answer a question, it's using their API.
API Key
A secret code (similar to a password) that gives a piece of software permission to access an AI service. Open Claw uses your API key to communicate with Claude, GPT, or other AI providers. Guard these like a password. Someone with your key can run up charges in your name.
C
Claw Chain SECURITY
A sequence of four linked security flaws discovered in May 2026 and named by Cyera (a cybersecurity firm). An attacker can exploit them in sequence — each step gives them more access — to eventually take complete control of an Open Claw installation. Dangerous because each step looks like normal agent activity. All four are fixed in the June update.
ClawHub ⚠ RISK
Open Claw's skill marketplace — the equivalent of an "app store" where you download additional capabilities for your agent. Contains 14,706+ skills. Google's cybersecurity team has confirmed it's an active target for malware distribution. Always verify before installing anything.
Claude
The AI made by Anthropic (a US AI safety company). Claude is one of many AI models that Open Claw can connect to as its "brain." Important: Open Claw and Claude are different things. Open Claw is the agent framework; Claude is one of the AI services it can use. You can also use ChatGPT (OpenAI), Gemini (Google), or others.
Container / Docker Sandbox
A way of running software in complete isolation. Think of it as a locked room — software inside the container can't see or affect anything outside it. If a malicious skill activates inside the container, it's trapped there. Docker is the most widely used tool for creating these containers. NanoClaw uses containers for every agent and skill by default.
CVE (Common Vulnerabilities and Exposures)
A standardized naming system for security flaws in software. Each CVE has a unique number (like CVE-2026-32922) so security teams worldwide can discuss the exact same flaw precisely. Think of it as the software industry's incident report number system. When you see a CVE number, it means the flaw has been officially cataloged.
CVSS Score
The "severity rating" for a security flaw on a scale of 0.0 to 10.0. A score of 9.9 means the flaw is critically serious — an attacker can likely take complete control of the affected system. A score of 5.0 is moderate. Most everyday software bugs score under 4. Two Open Claw flaws have scored 9.9 in 2026.
G
GitHub
The world's largest platform for storing, sharing, and collaborating on software code. When we say Open Claw has "375,000 GitHub stars," we mean 375,000 developers have bookmarked the project to show interest or support.
GitHub Stars
A way for developers on GitHub to bookmark or publicly endorse a project. High star counts indicate widespread community interest. It's a rough signal of popularity and credibility in the developer community.
Go (Programming Language)
A programming language created by Google, known for being fast and efficient. PicoClaw is written in Go, which allows it to run on tiny, cheap hardware.
GPU (Graphics Processing Unit)
A specialized computer chip originally designed for video games, now widely used for AI. NVIDIA's products (including DGX stations) are GPUs. NemoClaw is optimized to run AI models locally on NVIDIA GPU hardware.
GTIG (Google Threat Intelligence Group)
Google's team of professional cybersecurity researchers who track malware campaigns, hacking groups, and software vulnerabilities worldwide. In May 2026 they formally classified ClawHub as an active malware-distribution target.
I
IoT (Internet of Things)
Physical devices connected to the internet: smart thermostats, sensors, routers, cameras, vehicle computers. "Edge" bots in this guide (PicoClaw, Zero Claw, Null Claw) are specifically designed to run on IoT hardware.
M
MCP (Model Context Protocol)
A standard protocol that lets AI agents connect to external tools and applications — your calendar, task manager, company database, etc. Think of it as a universal connector for AI. Several bots in this guide support MCP, which dramatically expands what they can connect to.
MIT License
A type of open-source license that allows anyone to use, copy, modify, and share the software for free, including for commercial purposes. All the self-hosted bots in this guide are MIT licensed.
P
Port (Network Port) ⚠ RISK
A numbered "door" on a computer that allows incoming network connections. Open Claw's default is port 3000. Leaving this port exposed to the internet without protection is a major security risk. It's like leaving a door unlocked on the public street.
Prompt Injection ⚠ RISK
A type of attack where malicious text hidden in a document, email, or webpage tricks an AI agent into doing something unintended — like forwarding private files, sending messages, or changing settings. One of the harder attacks to defend against because it exploits how the AI understands language.
R
RAG (Retrieval-Augmented Generation)
A technique that makes AI look up answers in documents you provide, rather than guessing from its training data alone. Anything LLM uses RAG so you can ask questions of your own files and get cited, accurate answers. The "retrieval" part finds relevant passages; the "generation" part writes the answer.
Rust (Programming Language)
A programming language designed for high performance and built-in memory safety. It's structured so that a whole class of security vulnerabilities (called memory errors) are impossible by design. Zero Claw, Moltis, Open Fang, and IronClaw are written in Rust.
S
Self-Hosted
Running software on your own computer or a server you control, rather than using a cloud service. Self-hosted gives you full control over your data and configuration, but you're also responsible for security, updates, and uptime.
Skill ⚠ VERIFY FIRST
An add-on capability for Open Claw downloaded from ClawHub (the skill marketplace). Skills add new abilities — like reading email, monitoring a website, or connecting to Shopify. They're similar to browser extensions or app store apps. Current count: 14,706+, with 1,103 confirmed as malicious.
SQLite
A lightweight database that stores structured data in a single file on your computer or server. Several bots use SQLite to remember preferences and conversation history locally. No separate database server required.
SSH (Secure Shell)
A method for securely connecting to and controlling a remote server through a text interface. Required for advanced server management in Tier 1 self-hosting.
T
Tailscale
A free, easy-to-use private networking tool that creates a secure encrypted connection between your devices. Recommended for accessing your self-hosted Open Claw remotely instead of opening a network port to the internet.
Terminal / Command Line
A text-based interface for controlling a computer — you type commands instead of clicking. Most Tier 1 self-hosted bots require some comfort with the terminal for installation and configuration.
Token
The unit AI companies use to measure and charge for text processing. Roughly speaking, one token ≈ three-quarters of a word in English. "Per million tokens" is the standard pricing unit for AI APIs.
TypeScript
A programming language used by NanoClaw. It's a more structured and type-safe version of JavaScript (the language that powers most websites), widely used in professional web development.
V
VirusTotal
A free online service (owned by Google) that scans files and URLs using 70+ security scanners simultaneously. ClawHub now runs all uploaded skills through VirusTotal to check for known malware. A "Benign" result is a good sign, but not a guarantee.
VPS (Virtual Private Server)
A rented slice of a server in a data center, typically costing $6–20/month. The most common way to run a self-hosted Open Claw instance without owning your own hardware.
W
WASM / WebAssembly
A highly secure, efficient way of running code inside a tightly controlled sandbox environment. IronClaw uses WebAssembly to ensure the agent can only access exactly the resources you've explicitly permitted, with every access logged and auditable.
Webhook
An automatic notification sent from one app to another when something happens. For example: when a new email arrives, your email provider can send a webhook to Open Claw to process it immediately, rather than having the agent check for new email on a schedule.