AI Glossary: 176 terms in plain English

Short, accurate definitions of the terms you meet building with LLMs, RAG, agents and fine-tuning. Each entry links to the full explainer.

A

Adversarial Prompts: An adversarial prompt is any input deliberately crafted to make a language model fail. "Fail" is broad on purpose: the goal might be a wrong answer, a refusal where it should help, leaking its hidden instructions, ignoring its formatting rules, or breaking a safety policy. Read the full explainer →
Agent Framework: An agent framework is a library that handles the repetitive plumbing of building an AI agent, so you can focus on what the agent should actually do. Read the full explainer →
Agent Planning: Agent planning is the part of an AI agent that takes a big, fuzzy goal and turns it into a concrete, ordered list of smaller steps it can actually do. Read the full explainer →
AGENTS.md and CLAUDE.md: AGENTS.md and CLAUDE.md are the solution: plain Markdown files you drop into your repository that the agent reads before it does anything, giving it the standing briefing it needs to operate intelligently on your code. Read the full explainer →
Agno: Agno is a full-stack platform for building and running AI agents. Read the full explainer →
AI Agent: An AI agent is a large language model (LLM) that doesn't just answer you — it acts. Read the full explainer →
AI Alignment: AI alignment is the problem of getting an AI system to actually do what you intend — not just what you literally asked, and not just what scores well on a test. Read the full explainer →
AI Avatars: An AI avatar is a video of a person's face — real or synthetic — speaking words it was never recorded saying. Read the full explainer →
AI Coding Assistant: An AI coding assistant is a tool, built on a large language model (LLM), that helps you write and change code. Read the full explainer →
AI Engineer: An AI engineer is a software engineer who builds products on top of existing AI models instead of training new ones from scratch. Read the full explainer →
AI Red Teaming: AI red teaming is the practice of deliberately attacking your own AI system to find the ways it breaks — before a real user or an attacker finds them for you. Read the full explainer →
API Keys: An API key is a long, randomly generated string of characters — like sk-ant-api03-... or sk-proj-... — that you include in every request you send to an LLM provider. Read the full explainer →
ARC-AGI: ARC-AGI — the Abstraction and Reasoning Corpus — is a benchmark that tests something most AI tests don't: the ability to figure out a brand-new rule on the spot. Read the full explainer →
Argilla: Argilla is the open-source tool built for exactly that job. Read the full explainer →
Artificial Analysis: Artificial Analysis is an independent website that does that comparison for you. Read the full explainer →
assistant-ui: assistant-ui is an open-source React component library that hands you all of that plumbing. Read the full explainer →
AutoGen: AutoGen is a Microsoft Research open-source framework built around one deceptively simple idea: the best way to get AI agents to solve hard problems is to let them talk to each other. Read the full explainer →
Automatic Prompt Optimization: Automatic prompt optimization is that same loop, but a program drives it instead of you. Read the full explainer →
Autoregressive Generation: Autoregressive generation is how almost every large language model writes: it produces text one token at a time, and every token it writes gets fed right back in as part of the input for the next token. Read the full explainer →
Axolotl: Axolotl is an open-source tool that turns all of that into a single configuration file. Read the full explainer →

B

Benchmark Overfitting: Benchmark overfitting is what happens when a model gets really good at the test itself instead of the ability the test was meant to measure. Read the full explainer →

C

Chatbot Arena (LMSYS): Chatbot Arena is a website where you type a question, get two anonymous AI responses side by side, and vote for the one you prefer. Read the full explainer →
Chonkie: Chonkie is a small, fast open-source Python library that does exactly this one job well: it takes raw text and splits it into clean chunks, using whichever strategy fits your data. Read the full explainer →
Chroma: Chroma (also written ChromaDB) is an open-source vector store that runs directly inside your Python process. Read the full explainer →
Claude Agent SDK: The Claude Agent SDK is a software library from Anthropic that lets you build an autonomous AI agent in a few lines of code. Read the full explainer →
Claude API: The Claude API is how your code talks to Claude, Anthropic's family of AI models. Read the full explainer →
ComfyUI: ComfyUI is a free, open-source program for running AI image models on your own computer. Read the full explainer →
Computer Use Agents: A computer use agent is an AI that operates a computer the way a person does — it looks at the screen, decides what to click or type, takes the action, and checks the result. Read the full explainer →
Constitutional AI: Constitutional AI (CAI) is an alignment technique developed by Anthropic that teaches a language model to evaluate and rewrite its own outputs against a written list of principles — a "constitution" — rather than relying entirely on human raters to vote on which answers are good or bad. Read the full explainer →
Content Moderation for LLM Apps: Content moderation for LLM apps is the practice of running text — and sometimes images — through a safety check that flags or blocks harmful content before it enters the model or leaves it. Read the full explainer →
Context Compaction: Context compaction is the technique of automatically summarizing the older portion of that history into a compact form, then replacing the raw transcript with the summary. Read the full explainer →
Context Engineering: Context engineering is the discipline of packing that folder. Read the full explainer →
Context Window: A context window is the maximum amount of text a large language model can see and work with at one time. Read the full explainer →
Continued Pretraining: Continued pretraining (also called domain-adaptive pretraining or DAPT) is the process of taking a general-purpose language model that was trained on a broad internet corpus and running it through another round of the same pretraining objective — but now feeding it a large body of domain-specific text. Read the full explainer →
Continuous Batching: Continuous batching is the scheduling technique that allows a single GPU to serve many users simultaneously by treating the model's batch as a living, breathing queue rather than a locked room. Read the full explainer →
CopilotKit: CopilotKit is an open-source frontend stack for React and Angular that gives you all of that plumbing as ready-made building blocks. Read the full explainer →
CrewAI: CrewAI is an open-source Python framework that organizes AI agents the way a company organizes employees. Read the full explainer →
Cursor: Cursor is a code editor — the program you type your code into every day — rebuilt from scratch around AI. Read the full explainer →

D

DeepEval: DeepEval is an open-source framework that brings the unit-test feeling back to LLM work. Read the full explainer →
Diffusion Model: A diffusion model is the kind of AI that turns a text prompt like "a fox in a spacesuit" into a brand-new picture. Read the full explainer →
Dify: Dify is an open-source, low-code platform that bundles all of that into one place. Read the full explainer →
Direct Preference Optimization (DPO): Direct Preference Optimization (DPO) is a technique for teaching a language model to behave the way humans prefer — without building a separate reward model and without running a reinforcement-learning loop. Read the full explainer →
Distilabel: Distilabel is an open-source Python framework from Argilla for building synthetic-data and AI-feedback pipelines. Read the full explainer →
DistillKit: DistillKit is an open-source toolkit from Arcee AI for doing knowledge distillation on language models. Read the full explainer →

E

Embeddings: An embedding is a list of numbers that stands in for a piece of meaning. Read the full explainer →

F

FAISS: FAISS (Facebook AI Similarity Search) is an open-source library from Meta's Fundamental AI Research group. Read the full explainer →
FastAPI: FastAPI is the layer that turns your Python AI code into exactly that: a web service other programs can talk to. Read the full explainer →
FastEmbed: FastEmbed is a lightweight Python library from Qdrant that does exactly that one job — embeddings (and reranking) — without dragging in PyTorch. Read the full explainer →
FlashAttention: FlashAttention is a faster way to run the attention step inside a transformer. Read the full explainer →
FLUX: FLUX is a family of text-to-image AI models built by Black Forest Labs, a German startup founded by some of the original researchers behind Stable Diffusion. Read the full explainer →
Foundation Model: A foundation model is a massive neural network trained on a huge chunk of the internet — books, code, articles, forum posts — for the sole purpose of learning to predict the next token in a sequence. Read the full explainer →
Function Calling: Function calling is how a language model asks your program to do something it can't do on its own. Read the full explainer →

G

G-Eval: G-Eval is a recipe for building exactly that grader using another LLM. Read the full explainer →
Garak: Garak is an open-source tool from NVIDIA that does this testing for you, automatically. Read the full explainer →
Gemini API: The Gemini API is Google's public interface for running Gemini language models from your own code. Read the full explainer →
Generative UI: Generative UI is the pattern where the model's output drives which interface component your app renders, not just what words appear. Read the full explainer →
GitHub Copilot: GitHub Copilot is an AI assistant built into your code editor. Read the full explainer →
Golden Dataset: A golden dataset is a curated collection of input-output pairs that you trust. Read the full explainer →
Google ADK: Google ADK (Agent Development Kit) is an open-source, code-first framework for building AI agents and multi-agent systems. Read the full explainer →
Google Colab: Google Colab (short for Colaboratory) is a free service that gives you a working Python coding environment inside your web browser. Read the full explainer →
Google Veo: Google Veo is a text-to-video model from Google DeepMind. Read the full explainer →
GPTCache: GPTCache is an open-source library that sits in front of your LLM and remembers past answers. Read the full explainer →
Gradio: Gradio is a Python library, maintained by Hugging Face, that wraps your function in a small web app — usually in just a few lines of code. Read the full explainer →
GRPO: GRPO — Group Relative Policy Optimization — is a way to train a language model with reinforcement learning so it gets better at tasks where you can check whether the answer is right. Read the full explainer →
Guardrails AI: Guardrails AI is an open-source Python framework that puts a checkpoint around your LLM calls. Read the full explainer →

H

Haystack: Haystack is an open-source Python framework from deepset for building production-ready applications powered by large language models. Read the full explainer →
Hosted Fine-Tuning: Hosted fine-tuning means you customize a language model by uploading a data file to a cloud API — no GPU rental, no CUDA setup, no training loop to write. Read the full explainer →
Hugging Face: Hugging Face is a website and a set of tools that became the central home for open AI models. Read the full explainer →

I

Inference Server: An inference server is a program that loads a large language model onto a GPU once and then answers a constant stream of requests from many users at the same time — fast, and without falling over. Read the full explainer →
Inspect: Inspect is an open-source framework for testing AI models — built and maintained by the UK AI Security Institute (AISI). Read the full explainer →
Instructor: Instructor is a small Python library (also available in TypeScript, Go, Ruby, and more) that wraps your existing LLM client and adds one superpower: the response comes back as a validated Pydantic model, not a raw string. Read the full explainer →

J

Jailbreak: A jailbreak is a carefully crafted prompt — or a sequence of prompts — that convinces a large language model to ignore the safety rules it was trained to follow. Read the full explainer →
Jan: Jan is a free, open-source desktop app that flips this around. Read the full explainer →

K

Kling: Kling is an AI model that turns words or a single picture into a short video clip. Read the full explainer →
Knowledge Cutoff: A knowledge cutoff is the date after which a language model has, in effect, read nothing. Read the full explainer →
Kokoro TTS: Kokoro TTS is a text-to-speech model: you feed it written text, and it produces a spoken-audio waveform you can play back. Read the full explainer →

L

LanceDB: LanceDB is an open-source vector database that runs inside your application — no server to install, no service to keep alive. Read the full explainer →
LangChain: LangChain is an open-source toolkit for building apps powered by large language models (LLMs). Read the full explainer →
Langfuse: Langfuse is an open-source observability platform built for exactly this. Read the full explainer →
LangGraph: LangGraph is a low-level orchestration library from the LangChain team that lets you build agents as explicit graphs. Read the full explainer →
LangSmith: LangSmith is an observability and evaluation platform built by the LangChain team. Read the full explainer →
Large Language Model (LLM): A large language model (LLM) is a computer program trained to do one deceptively simple thing: look at a piece of text and predict what comes next. Read the full explainer →
Least-to-Most Prompting: Least-to-most prompting is a way of getting a language model to solve a hard problem by first breaking it into a ladder of smaller subproblems, then solving them one at a time — easiest first — feeding each answer into the next step. Read the full explainer →
LiteLLM: LiteLLM is the universal adapter that makes all of them speak one language. Read the full explainer →
LLaMA-Factory: LLaMA-Factory is a toolkit that turns all of that plumbing into a handful of choices. Read the full explainer →
LlamaIndex: LlamaIndex is an open-source Python (and TypeScript) framework that connects your own data — PDFs, Notion pages, a database, a folder of Word docs — to a large language model so the model can answer questions about it. Read the full explainer →
LLM API: An LLM API is the doorway your code uses to talk to an AI model that lives on someone else's servers. Read the full explainer →
LLM Benchmarks: An LLM benchmark is a standardized test for an AI model. Read the full explainer →
LLM Evals: An LLM eval (short for evaluation) is a repeatable test for an AI's output. Read the full explainer →
LLM Gateway: An LLM gateway (also called an LLM proxy) is a small server you run between your apps and the providers. Read the full explainer →
LLM Guard: LLM Guard is an open-source security toolkit, built by Protect AI, that sits around your model and inspects both streams. Read the full explainer →
LLM Guardrails: LLM guardrails are the safety checks that run around a model call — one set on the way in, another on the way out. Read the full explainer →
LLM-as-a-Judge: LLM-as-a-judge means using one language model to grade the output of another. Read the full explainer →
LLMOps: LLMOps is the practice of running an LLM-powered app in production and keeping it healthy once real users are hitting it. Read the full explainer →
lm-evaluation-harness: lm-evaluation-harness is an open-source tool from EleutherAI that runs academic benchmarks on a language model for you. Read the full explainer →
LMArena: LMArena (now rebranded to simply Arena at arena.ai) is the internet's biggest live tournament for AI chatbots. Read the full explainer →

M

MCP (Model Context Protocol): MCP (Model Context Protocol) is an open standard for plugging tools and data sources into AI apps. Read the full explainer →
Mem0: Mem0 is a memory layer you bolt onto an agent so it stops forgetting. Read the full explainer →
Midjourney: Midjourney is an AI image generator: you type a description in plain language, and it paints a picture to match. "A lighthouse at dusk in heavy fog, oil painting" goes in; a few finished images come back. Read the full explainer →
Milvus: Milvus is an open-source vector database designed to store and search huge numbers of embeddings — the lists of numbers that AI models produce to represent the meaning of text, images, or audio. Read the full explainer →
MLX: MLX is Apple's open-source framework for running and training machine learning models on a Mac. Read the full explainer →
Model Distillation: Model distillation is how you take a huge, slow, expensive language model and bottle its behavior into a tiny, fast, cheap one. Read the full explainer →
Model Routing: Model routing is the practice of automatically deciding, for each incoming request, which AI model should handle it — and defaulting to the cheapest model that can do the job well enough. Read the full explainer →
Multi-Agent System: A multi-agent system is a setup where several AI agents work on one job together, each handling a piece of it, instead of a single agent doing everything alone. Read the full explainer →

N

n8n: n8n (pronounced "n-eight-n," short for nodenation) is a tool for building automations by dragging boxes onto a canvas and wiring them together with lines. Read the full explainer →
Nano Banana: Nano Banana is the nickname for Google's Gemini Flash Image model — the part of Gemini that can make and edit pictures, not just describe them. Read the full explainer →
Negative Prompts: A negative prompt is a second, separate prompt that lists what you don't want: blurry, extra legs, watermark, text. Read the full explainer →
NeMo Curator: NeMo Curator is NVIDIA's open-source framework for doing exactly that job at massive scale. Read the full explainer →
Next-Token Prediction: Next-token prediction is the single trick behind every large language model. Read the full explainer →

O

Ollama: Ollama is a free, open-source tool that lets you download and run large language models on your own computer with a single command. Read the full explainer →
OpenAI Agents SDK: The OpenAI Agents SDK is an open-source Python (and TypeScript) library that gives you a tiny, production-ready framework for wiring AI agents together. Read the full explainer →
OpenAI API: The OpenAI API is how your code talks directly to OpenAI's GPT models — the current GPT-5 series and the rest. Read the full explainer →
OpenAI Realtime API: The OpenAI Realtime API is a way to build a voice assistant that you can talk to like a person — you speak, it listens, it thinks, and it speaks back, all in the same conversation. Read the full explainer →
OpenRLHF: OpenRLHF is an open-source framework for running RLHF — reinforcement learning from human feedback — on large language models, at a scale that doesn't fit on a single GPU. Read the full explainer →
OpenRouter: OpenRouter is a single front door that stands in front of all of them. Read the full explainer →
Opik: Opik (made by Comet, and released under the permissive Apache-2.0 license) is an open-source tool that gives you that missing record. Read the full explainer →
Orchestrator-Worker Pattern: The orchestrator-worker pattern is a way of organizing multiple AI agents where one lead agent (the orchestrator) breaks a job into sub-tasks and hands each sub-task to a worker agent. Read the full explainer →

P

PaddleOCR: PaddleOCR is a mature, free, open-source toolkit that does this whole job for you. Read the full explainer →
PEFT: PEFT stands for parameter-efficient fine-tuning, and Hugging Face PEFT is the open-source Python library that makes it easy. Read the full explainer →
pgvector: pgvector is a free, open-source extension for PostgreSQL that turns your existing Postgres database into a vector search engine. Read the full explainer →
Portkey: Portkey is an AI gateway: a single layer that sits between your application and every model provider you call — OpenAI, Anthropic, Google, your own self-hosted models — and turns the messy job of running LLMs in production into something you can see, control, and govern from one place. Read the full explainer →
Prompt Compression: Prompt compression is the practice of shrinking that input — cutting filler, summarizing old material, and removing duplicates — so the model receives the same useful information in far fewer tokens. Read the full explainer →
Prompt Engineering: Prompt engineering is the craft of writing instructions for an AI model so that it reliably does what you actually want. Read the full explainer →
Prompt Management: Prompt management is the practice of treating that text like real source code: it lives in one known place, every edit creates a new numbered version, changes are reviewed and tested before users see them, and when a change goes wrong you can roll back in seconds. Read the full explainer →
Promptfoo: Promptfoo is an open-source command-line tool that turns this eyeballing into a real test suite. Read the full explainer →
PydanticAI: PydanticAI is a Python agent framework built by the same team behind Pydantic — the validation library that OpenAI, Anthropic, and Google all use internally. Read the full explainer →

Q

Qdrant: Qdrant (pronounced "quadrant") is an open-source vector database and similarity-search engine written in Rust. Read the full explainer →
QLoRA: Unsloth is a popular open-source library (2024–2025) that rewrites the QLoRA forward and backward passes with custom CUDA kernels. Read the full explainer →
Qwen-VL: Qwen-VL is Alibaba's open-weight family of vision-language models (VLMs) that fixes exactly that: it takes images and video and text as input, and reasons over them together to produce a text answer. Read the full explainer →

R

Ragas: Ragas is an open-source Python library that grades a RAG or LLM application automatically. Read the full explainer →
RAGFlow: RAGFlow is an open-source, end-to-end RAG engine built around one idea: understand the document before you chunk it. Read the full explainer →
Reward Model: A reward model is a small AI model with one narrow job: read a piece of text an LLM produced and return a single number that says how good it is. Read the full explainer →
Runway: Runway is an online studio that turns a text prompt or a still image into a short video clip. Read the full explainer →

S

Schema Validation for LLM Output: Schema validation for LLM output is the practice of taking whatever text a model returns and running it through a typed shape definition — a schema — that says exactly what fields must be present, what types they must be, and what values are acceptable. Read the full explainer →
Semantic Caching: Semantic caching is a way to reuse an LLM's answer when a new question means the same thing as one you've already answered — even if the wording is completely different. Read the full explainer →
Semantic Kernel: Semantic Kernel is an open-source SDK from Microsoft for building applications powered by large language models. Read the full explainer →
sentence-transformers: sentence-transformers (also called SBERT) is the open-source Python library that does exactly that. Read the full explainer →
SGLang: SGLang is an open-source framework for serving large language models — the software that sits between your application and a GPU, taking in requests and streaming back tokens as fast as the hardware allows. Read the full explainer →
Sliding Window Attention: Sliding window attention is a simple, brutal fix: don't let each token look at everything — only at a fixed number of recent neighbors. Read the full explainer →
smolagents: smolagents is a tiny open-source agent library from Hugging Face. Read the full explainer →
Spec-Driven Development: Spec-driven development (SDD) is a workflow where you write a structured document describing what you want to build before you ask an AI agent to write a single line of code. Read the full explainer →
Specification Gaming: Specification gaming is that genie, except the genie is an AI system and the wish is the goal you gave it. Read the full explainer →
Speculative Decoding: Speculative decoding is a trick to skip a lot of that waiting without changing the answer at all. Read the full explainer →
Speech-to-Text: Speech-to-text is exactly what it sounds like: you give a computer a recording of someone talking, and it hands back the words they said as plain text. Read the full explainer →
SPLADE: SPLADE is a way to search text that combines the speed of old keyword search with the smarts of a neural network. Read the full explainer →
Spring AI: Spring AI is the official Spring project that does exactly that. Read the full explainer →
sqlite-vec: sqlite-vec is a tiny extension that teaches that same little file a new trick: storing embeddings and searching them by meaning — the core job of a vector database. Read the full explainer →
Stable Diffusion: Stable Diffusion is a family of open-weight text-to-image models from Stability AI. Read the full explainer →
Stanford HELM: Stanford HELM — short for Holistic Evaluation of Language Models — is a framework built on a simple idea: stop judging a model by one number, and measure it across many aspects at once. Read the full explainer →
Step-Back Prompting: Step-back prompting is a simple two-step trick: before you ask a model to solve a specific problem, you first ask it to step back and name the general principle, concept, or rule that the problem is really about. Read the full explainer →
Stop Sequences: A stop sequence is a piece of text you hand the API in advance that says: "the instant you produce this exact string, stop — don't write another token." Read the full explainer →
Strands Agents: Strands Agents is an open-source Python SDK from AWS for building AI agents — programs that don't just answer a question once, but loop: think, call a tool, read the result, think again, and keep going until the task is done. Read the full explainer →
Subagents: A subagent is a fresh AI agent that a main agent launches to handle one specific subtask, then throws away when the subtask is done. Read the full explainer →
Supabase: Supabase is an open-source backend-as-a-service built on top of PostgreSQL, one of the most trusted relational databases in the world. Read the full explainer →
SWE-bench: Each SWE-bench task instance is built from one real merged pull request. Read the full explainer →
Synthetic Training Data: Synthetic training data is training data written by a model instead of a human. Read the full explainer →
System Prompt: A system prompt is the set of standing instructions a developer gives an AI model before any user types a word. Read the full explainer →

T

Tambo: Tambo is an open-source generative-UI SDK for React that lets your agent answer with actual React components instead of only words. Read the full explainer →
TensorRT-LLM: TensorRT-LLM is NVIDIA's library for building that better plumbing. Read the full explainer →
Text-to-Speech: Text-to-speech (TTS) is software that reads written text out loud. Read the full explainer →
Thesys C1: Thesys C1 is a commercial, hosted platform that tackles exactly this gap. Read the full explainer →
tiktoken: tiktoken is OpenAI's fast, open-source library that turns text into exactly the same tokens an OpenAI model would see — before you send anything over the network. Read the full explainer →
Time to First Token: Time to First Token (TTFT) is the duration of that silence: the milliseconds between the moment your request leaves the client and the moment the very first character of the response arrives. Read the full explainer →
TRL: TRL (Transformer Reinforcement Learning) is the open-source library from Hugging Face that packages all of that into ready-made Trainer classes. Read the full explainer →
txtai: txtai is an open-source Python library that rolls all of that into one package. Read the full explainer →

U

Unsloth: Unsloth is an open-source library that makes that same fine-tuning run faster and fit in far less VRAM, so a job that once needed a rented data-center card can often run on a single consumer GPU or even a free cloud notebook. Read the full explainer →
Unstructured: Unstructured is an open-source library that fixes the very first step. Read the full explainer →

V

Vercel AI SDK: The Vercel AI SDK is a TypeScript toolkit for talking to AI models from your own code. Read the full explainer →
verl: verl is an open-source framework for doing reinforcement-learning post-training of large language models at scale. Read the full explainer →
Vibe Coding: Vibe coding is a style of software development where you describe what you want in plain language, let an AI generate the code, and accept it without reading it carefully — trusting the "vibe" of the session rather than verifying each line. Read the full explainer →
vLLM: vLLM is an open-source inference server built at UC Berkeley that became the de-facto standard for high-throughput LLM serving. Read the full explainer →

W

Wan: Wan is a family of open-source AI video models from Alibaba. Read the full explainer →
Weaviate: Weaviate's module system is what separates it from lower-level vector stores. Read the full explainer →
Whisper: Whisper is OpenAI's open-source speech-to-text model: you hand it an audio file of someone talking, and it hands back the words as text. Read the full explainer →
World Model: A world model is an AI system that learns an internal simulation of how some environment behaves — how it looks, and crucially how it changes when something acts on it. Read the full explainer →

X

XGrammar: XGrammar is one of the engines that does this, and does it fast. Read the full explainer →