Glossary · GenAI Field Guide

The model

Term	One line
LLM	A model that predicts the next token; fluent, confident, memoryless.
Token	About ¾ of a word; the unit everything is measured in.
Reasoning / thinking mode	A setting where the model works through steps before answering; slower, better for hard multi-step tasks. Sometimes set as a level or a token budget.
Multimodal	A model that reads images, documents, audio, or video as well as text.
Hallucination	Confident, fluent, wrong; a structural property, not a bug to be patched away.
Parameters (7B, 35B, 70B)	The learned numbers inside a model, counted in billions; a rough size, not a quality score.
Precision / quantization	How many bits store each parameter; quantizing (Q8, Q4) shrinks and speeds a model, with a quality cost that is small at 8-bit and can be real at 4-bit.
Open-weight vs closed-weight	Open-weight models publish their parameters, so anyone can run them; closed-weight models are reached only through the vendor's service. Whether a model is hosted for you or run on your own hardware is a separate question.
Knowledge cutoff	The date a model's training data ends; anything later must come from search or your documents.
Tier (Flash, Pro, mini)	Size-and-speed variants of one model generation.
Frontier model	The newest, largest models at the top of a vendor's line.
Benchmark	A score on a public test, usually reported by the vendor; a rough ranking, not your task, and marketing until independently reproduced.
Fine-tuning	Further training a model on your examples so it speaks differently; rarely the right first tool.

The prompt

Term	One line
Prompt	The entire text the model receives this turn, not just what you typed.
System prompt / preamble	Vendor and organization instructions placed before your message.
Context window	The maximum tokens per call; your conversation shares it with the preamble, tools, and files.
Compaction	Summarizing older turns to make room; detail is lost.
Harness	The product layer around the model: preamble, tools, memory, filters, compaction.
Project	Persistent files plus instructions across chats; RAG-lite.
Prompt injection	Instructions hidden in data (a document, a web page) that steer the model as if you typed them.
Impact Level (IL)	DoD cloud data-sensitivity tier; IL5 = CUI, not classified.

Agents and tools

Term	One line
Tool	A function the model can request; the harness runs it and returns text.
Tool result	That returned text, appended to the context (including program output).
Schema	The fixed shape of a tool's inputs or an output (fields, types); what a program can check.
MCP	Model Context Protocol, the standard connector format for tools; "connectors" are MCP servers.
Skill	A loadable instruction set (plus scripts) for a repeatable task.
Agent	An LLM in a loop with tools, until a stop condition.
Agentic	Able to take multiple steps and use tools on its own before reporting back.
Subagent	A fresh-context helper spawned by an agent; returns only a summary.
Session	One running transcript.
RAG	Retrieve relevant chunks of your documents, then generate from them.
Embedding	A vector that encodes meaning; how RAG finds "relevant."
API	Calling the model from code; full control, none of the harness's safety rails.
Deterministic	Gives the same answer every time for the same input: a calculator, a spreadsheet, a script. The opposite of a language model.