Dictionary
A small AI vocabulary
Buzzwords show up in meetings faster than definitions do. These pages keep a Quiet Nodes meaning for the terms we actually use in the cups — short, linkable, and honest about what a word does not mean.
- Agent Software that uses a model to decide the next step, then acts (often with tools), instead of only answering one prompt.
- API A doorway between programs: one piece of software asks another for data or an action through an agreed format.
- CLI Command-line interface: talking to software by typing in a terminal instead of clicking through a graphical UI.
- Context window How much text a model can keep in working memory for one reply: your prompt, history, documents, and its answer share the same budget.
- Harness The thin (or not so thin) software wrapped around a model: tools, UI, memory, guardrails — where product value often actually sits.
- MCP Model Context Protocol: a shared way for AI apps to plug into tools and data, so every pair does not need its own adapter.
- Prompt The text (or speech) you give a model: instructions, questions, and whatever context you want it to use.
- RAG Retrieval-augmented generation: fetch relevant notes or docs first, then let the model answer with that material in view.
- SLM Small language model: a lighter model that can run locally or cheaply for focused jobs, instead of waking a giant every time.
- Tokens The little chunks models read and write; billing, context limits, and ‘how long is this prompt’ all count in tokens.