An AI agent is a program built on a language model that completes tasks on its own: it understands a request, plans steps, calls tools, and responds with a result instead of just text.

The context window is the maximum amount of text (measured in tokens) a language model can process at once — typically 128k to 1M tokens with current models.

A Data Processing Agreement (DPA) is the legal contract that governs how a provider processes personal data on behalf of the customer — GDPR-mandatory for any AI vendor.

An embedding is a numeric representation of text (or an image) in a high-dimensional space where similar content sits close together — the foundation of semantic search and RAG.

Fine-tuning is the additional training of an already-trained language model on your own data — changes the model itself, unlike RAG.

A guardrail is a safety layer between user and language model that blocks unwanted inputs or outputs — e.g. PII leaks, prompt injection, or off-topic requests.

A hallucination is a plausible-sounding but factually wrong answer from a language model — e.g. a made-up quote or an invented software feature.

A Large Language Model is a neural network trained on billions of texts that understands, generates and transforms language. GPT-4, Claude, and Mistral are examples.

Model Context Protocol is an open standard from Anthropic that defines how AI agents talk to tools, data sources, and APIs — like USB for AI.

Multi-agent refers to an architecture where several specialized AI agents collaborate — a router decides who takes over, and they hand off tasks cleanly.

A prompt is the instruction a language model receives — it defines role, tone, allowed actions, and provides the context for the response.

RAG (Retrieval-Augmented Generation) is a technique where the language model retrieves relevant documents from your knowledge base before each answer and uses them as the basis for its response — so the agent stays current and answers with sources.

Streaming means the language model's response is delivered token by token as soon as it's generated — the user sees the beginning while the model is still writing.

Tool use (or function calling) is a language model's ability to call external functions with structured arguments — e.g. an API, a database query, or a skill.

A vector database stores embeddings efficiently and answers similarity searches in milliseconds — the storage behind every RAG system.