AI Glossary: Plain English for Business
Demystify AI terminology. No jargon. Just clear explanations for business leaders and technical teams.
A
Agent
An autonomous software system that observes its environment, makes decisions, and takes actions without continuous human instruction. Think of it as a helpful assistant that can run tasks independently. In PrivateClaw, agents orchestrate workflows across your business systems using N8N.
Intermediate
Attention Mechanism
The technique that lets AI models focus on the most relevant parts of input data (like a person paying attention to key words in a sentence). It’s the core technology behind modern language models like Claude and GPT.
Advanced
B
Batch Processing
Running multiple jobs together (a “batch”) instead of one at a time. Example: analysing 1,000 emails at once rather than one per request. More efficient for large-scale work.
Beginner
C
Claude
Anthropic’s AI assistant—a large language model (LLM) built for reasoning, coding, and deep analysis. Claude can be self-hosted via API or accessed via PrivateClaw’s orchestration layer for private inference.
Beginner
ComfyUI
An open-source, node-based interface for generative AI image models (like Stable Diffusion). Lets non-technical users build complex image workflows by connecting blocks visually. Included in PrivateClaw Community and Kick Start tiers.
Intermediate
Context Window
The amount of text an AI model can "see" at once. Claude has a 200K-token context window—roughly equivalent to 150,000 words. Larger context windows let models work with longer documents and files.
Intermediate
D
Docker
A containerisation platform that packages your software (and all its dependencies) into isolated, portable units. Think of it as a shipping container for code—runs the same way on any machine. Essential for deploying PrivateClaw infrastructure.
Intermediate
E
Embedding
A mathematical representation of text or data as a series of numbers. Embeddings capture meaning—similar texts have similar embeddings. Used for search, similarity matching, and as input to other AI models.
Intermediate
F
Fine-tuning
Training an AI model on your specific data so it learns your domain, style, and requirements. Like teaching an assistant to sound and behave like your brand. Computationally lighter than training from scratch.
Advanced
G
GPU (Graphics Processing Unit)
Specialised hardware designed to perform thousands of calculations in parallel. GPUs accelerate AI model inference and training dramatically compared to CPUs. PrivateClaw hardware recommendations centre on NVIDIA GPUs.
Beginner
H
Hallucination
When an AI model confidently generates false or made-up information. Like a person making up a fact instead of admitting they don’t know. Happens with all LLMs; combated through RAG (retrieval-augmented generation) and verification.
Intermediate
I
Inference
Running a trained AI model to generate predictions or outputs. Example: feeding text into Claude to get a response. Inference is fast compared to training; this is what end-users experience.
Intermediate
K
Knowledge Graph
A structured representation of entities and their relationships. Example: "John (Person) works at PrivateClaw (Company)". Used in RAG systems and AI agents to ground models in factual, contextual data.
Advanced
L
LLM (Large Language Model)
An AI model trained on vast amounts of text to predict and generate human language. Claude, GPT-4, and Llama are LLMs. They power chatbots, summarisation, coding, and reasoning tasks.
Beginner
LoRA (Low-Rank Adaptation)
A lightweight fine-tuning technique that adds small, trainable layers to a pre-trained model instead of modifying the whole model. Much cheaper and faster than full fine-tuning while achieving strong results.
Advanced
M
MCP (Model Context Protocol)
A standardised way for AI models to safely access external tools, databases, and APIs. Think of it as a contract between the model and your systems. PrivateClaw uses MCP for secure agent-to-system communication.
Advanced
Model
A trained AI system that takes input and produces output based on patterns learned from training data. A model is like a compiled program—it runs inference fast, but took time to build.
Beginner
N
N8N (n8n)
An open-source workflow automation platform. Build complex multi-step processes by connecting nodes visually—no coding required. PrivateClaw uses N8N as its orchestration backbone for agents and automations.
Intermediate
Neural Network
A computational structure inspired by biological brains, made of interconnected nodes (neurons). Neural networks learn by adjusting connection strengths during training. Foundation for modern deep learning and LLMs.
Intermediate
O
Ollama
A lightweight tool for downloading and running large language models locally on your own hardware. Supports Llama, Mistral, Deepseek, and 100+ open models. Core technology in PrivateClaw’s local-first architecture.
Intermediate
On-Premise (On-Prem)
Software and data hosted and run on your own servers, in your office or data centre—not on someone else’s cloud. Gives you full control and privacy. PrivateClaw is designed for on-premise deployment.
Beginner
OpenClaw
PrivateClaw’s AI orchestration engine for coordinating multi-model workflows, agent communication, and system integration. Sits atop Ollama, N8N, and your LLMs.
Advanced
P
Parameter
A learnable value inside an AI model. A model with 7 billion parameters has 7 billion numbers that were adjusted during training. More parameters generally mean more capability, but also more memory and compute needed.
Intermediate
Prompt Engineering
The art of writing effective instructions to get the best outputs from an AI model. Small changes in wording or structure can dramatically change results. A key skill for working with LLMs.
Beginner
Q
Quantisation
Reducing the precision of a model’s numbers (from 32-bit floats to 8-bit integers, for example) to make it smaller and faster. You lose a tiny bit of accuracy but gain massive speed and memory efficiency. Essential for running models locally.
Advanced
R
RAG (Retrieval-Augmented Generation)
A technique where an AI model retrieves relevant documents or facts before generating a response, grounding answers in your actual data. Reduces hallucinations and lets models use proprietary information.
Intermediate
S
SIEM (Security Information & Event Management)
A system that collects, analyses, and responds to security events across your infrastructure. Watches for threats, logs access, and triggers alerts. PrivateClaw Kick Start integrates SIEM for compliance and audit trails.
Advanced
T
Token
The smallest unit of text that an AI model processes. Usually a word or part of a word. Context windows and pricing are measured in tokens. 1,000 tokens ≈ 750 words.
Beginner
Transformer
The neural network architecture that powers modern LLMs. Uses attention mechanisms to process sequences of data in parallel, capturing long-range dependencies in text. Revolutionary for natural language.
Advanced
V
Vector Database
A database optimised for storing and searching embeddings. Lets you find semantically similar documents or images fast—foundational for RAG and vector search. Examples: Pinecone, Weaviate, Milvus.
Advanced
W
WSL2 (Windows Subsystem for Linux 2)
A layer that lets you run a real Linux kernel inside Windows, enabling Docker and Linux tools natively on Windows. PrivateClaw development and testing often uses WSL2 for portability.
Intermediate
Z
Zero-Shot Learning
Asking an AI model to do a task it has never explicitly trained on, just from a description. Example: asking Claude to summarise a document in a style it hasn’t been trained on specifically. Often works well with modern LLMs.
Intermediate
Still Confused?
Book a Foundation training session to build your AI vocabulary with instructors who actually care about clarity.