Glossary

A2A Protocol (Agent-to-Agent Protocol) is an open interoperability standard initiated by Google that...

Adapter is base-dependent trainable state that changes a pretrained model's behavior while most base...

AES (Advanced Encryption Standard) is a symmetric block cipher algorithm adopted by the U.S. governm...

Agent Development Kit is a framework layer that provides reusable abstractions for building AI agent...

Agent Harness is the application control system around an AI agent that assembles context, constrain...

Agent Memory is the governed storage and retrieval of information that an AI agent may use across st...

Agent Runtime is the execution environment that creates, schedules, isolates, advances, suspends, re...

Agent Trajectory is a structured, causally ordered record of the observable decisions, messages, too...

An Agentic Workflow is a business or technical process that delegates selected runtime decisions to ...

AGI (Artificial General Intelligence) is a theoretical form of artificial intelligence that possesse...

AI Accelerator is specialized hardware and its supporting software stack designed to execute machine...

AI Agent is an autonomous software system powered by Large Language Models, implementing goal-orient...

AI App Builder is a category of no-code/low-code platforms that use large language models to generat...

AI Code Review is the use of machine-learning or language-model systems to inspect proposed code cha...

AI Guardrails are versioned technical and operational controls that detect, constrain, block, transf...

An AI Watermark is a signal intentionally embedded in generated or manipulated content so a correspo...

Answer Faithfulness is a RAG evaluation property that measures whether the factual claims in a gener...

API (Application Programming Interface) is a set of rules, protocols, and tools that allows differen...

An API Gateway is a server component that acts as the single entry point for all client API requests...

Approval Gate is a deterministic enforcement checkpoint that blocks a proposed AI action until trust...

Artificial Intelligence (AI) is a branch of computer science that focuses on creating intelligent sy...

ASCII (American Standard Code for Information Interchange) is a character encoding standard that ass...

Aspect Ratio is the proportional relationship between the width and height of an image, video, or di...

Attention Mechanism is a differentiable operation that computes data-dependent weights over permitte...

AutoGen is Microsoft's open-source framework for building conversational agents and event-driven mul...

Autonomous Driving is a technology that enables vehicles to navigate and operate without human inter...

AWQ (Activation-aware Weight Quantization) is a post-training, weight-only quantization method that ...

Backpropagation is a fundamental algorithm for training artificial neural networks that efficiently ...

Barcode is a machine-readable representation of data using patterns of parallel lines (1D) or geomet...

Base64 is a binary-to-text encoding scheme that represents binary data in an ASCII string format by ...

Bearer Token is an access token type used in HTTP authentication where the client presents a token t...

Bi-Encoder is a retrieval model architecture that encodes queries and documents separately into embe...

Binary (also known as base-2) is a numeral system that uses only two symbols: 0 and 1. It is the fun...

BM25 is a probabilistic lexical ranking function that scores documents based on query term matches, ...

BMI is a numerical value calculated from a person's weight and height, used as a screening tool to c...

Callback Handler is a lifecycle hook component that receives events from model calls, tools, retriev...

Catastrophic Forgetting is the loss or degradation of previously learned capabilities when a model i...

CDN is a geographically distributed network of proxy servers and data centers that work together to ...

Chain Orchestration is an LLM application pattern in which code defines an ordered path of model, re...

Chain-of-Thought (CoT) is a prompting technique that elicits intermediate natural-language steps bef...

Chat Template is a model-specific serialization contract that converts structured messages, roles, t...

Chatbot is an artificial intelligence software application designed to simulate human-like conversat...

ChatModel is an application-layer abstraction for invoking conversational language models through st...

Checksum is a small-sized block of data derived from another block of digital data for the purpose o...

Chunk Overlap is the repeated text shared between adjacent document chunks so that information near ...

Chunk Size is the token, character, or structural length chosen for each document unit indexed in a ...

Chunking is the process of splitting long documents or data sources into smaller retrievable units t...

Citation is the attribution of an AI-generated claim to a specific source document, passage, URL, re...

CMYK is a subtractive color model used in color printing, where cyan, magenta, yellow, and black ink...

CNN (Convolutional Neural Network) is a class of deep neural networks specifically designed for proc...

Cold Start is the additional startup latency that occurs when a model service handles a request befo...

Computer Vision is a field of artificial intelligence that enables computers to interpret and unders...

Constrained Decoding is an inference technique that filters an LLM's allowed next tokens so the gene...

Context Budget is an application-level policy that allocates a model request's finite token capacity...

Context Caching is a cross-request inference optimization that reuses provider- or runtime-managed s...

Context Compression is a lossy transformation that reduces the context supplied to an LLM while pres...

Context Engineering is the design and operation of the information supplied to a language model for ...

Context Precision is a RAG evaluation metric that measures how much of the retrieved context is rele...

Context Recall is a RAG evaluation metric that measures whether the retrieved context contains the e...

Context Window is the maximum tokenized sequence a model endpoint accepts for one inference request ...

Continuous Batching is an LLM serving technique that dynamically groups active requests during infer...

CORS (Cross-Origin Resource Sharing) is a security mechanism that allows web browsers to make reques...

CrewAI is an open-source Python framework specifically designed to coordinate multiple autonomous Ag...

Cron is a time-based job scheduler in Unix-like operating systems that enables users to schedule com...

Cross-Encoder is a ranking model architecture that jointly encodes a query and a candidate document ...

CSS (Cascading Style Sheets) is a stylesheet language used to describe the presentation of HTML and ...

CSV (Comma-Separated Values) is a plain text file format that stores tabular data using commas to se...

Data URL is a URI scheme that allows embedding small data items inline in web documents as if they w...

Dataset Curation is the process of selecting, cleaning, organizing, labeling, deduplicating, and val...

Decode Phase is the LLM inference stage that generates output one token at a time using the KV cache...

Deep Learning is a subset of machine learning that trains multi-layer neural networks to fit represe...

Dense Retrieval is a semantic search method that represents queries and documents as dense embedding...

Diff is a comparison technique that identifies and displays the differences between two sets of data...

Diffusion Model is a family of generative models that learns a reverse denoising transition from noi...

Disaggregated Serving is an LLM inference architecture that runs prompt prefill and token decoding i...

Distillation (Knowledge Distillation) is a model compression technique where a smaller 'student' mod...

DiT (Diffusion Transformer) is a generative model architecture that replaces the traditional U-Net b...

DNS (Domain Name System) is a hierarchical and distributed naming system that translates human-reada...

Document Loader is an ingestion component that reads raw content from files, web pages, object stora...

Document Transformer is a pipeline component that cleans, splits, enriches, filters, or restructures...

DPO (Direct Preference Optimization) is a preference-learning objective that adjusts a language mode...

Eino is an open-source Go framework in the CloudWeGo ecosystem for building production-oriented LLM ...

Embedding is a learned numerical representation that maps text, images, entities, or other inputs in...

Emoji is a standardized set of pictographic symbols and icons encoded in Unicode, used in digital co...

Encryption is the process of converting plaintext data into an unreadable format (ciphertext) using ...

Epoch time refers to the reference point (January 1, 1970, 00:00:00 UTC) from which Unix timestamps ...

Escape Character is a character that invokes an alternative interpretation of subsequent characters ...

EXIF is a standard that specifies the formats for images, sound, and ancillary tags used by digital ...

Few-Shot Learning is a machine learning paradigm where models learn to perform tasks from only a sma...

Fine-tuning is the process of continuing optimization from a pretrained model checkpoint on a target...

Function Calling is a large-language-model capability that produces a structured proposal naming a t...

GAN (Generative Adversarial Network) is a class of deep learning models consisting of two neural net...

Generative AI is a category of artificial intelligence systems capable of creating new content—inclu...

Geohash is a geocoding system that encodes geographic coordinates (latitude and longitude) into a sh...

GIF (Graphics Interchange Format) is a bitmap image format developed by CompuServe in 1987 that supp...

Glob is a pattern matching syntax used to specify sets of filenames or paths using wildcard characte...

Go (also known as Golang) is a statically typed, compiled programming language designed at Google. I...

Golden Dataset is a versioned, human-reviewed evaluation set whose cases pair representative inputs ...

GPT (Generative Pre-trained Transformer) is a family of large language models developed by OpenAI th...

Gradient Descent is a first-order iterative optimization algorithm used to find the minimum of a fun...

Graph Orchestration is an execution pattern that represents an AI workflow as explicit nodes, transi...

GraphRAG (Graph Retrieval-Augmented Generation) is a RAG approach that represents entities and relat...

Grounding is the system practice of supplying a generative AI model with bounded external evidence a...

GRPO (Group Relative Policy Optimization) is an online policy-gradient method that estimates each sa...

GUID (Globally Unique Identifier) is Microsoft's implementation of the UUID standard, a 128-bit iden...

Gzip is a file compression format and software application that uses the DEFLATE algorithm to reduce...

Hallucination is confidently stated generated content that is erroneous or false relative to a decla...

Hash is a mathematical function that converts input data of any size into a fixed-size string of cha...

Hex Color is a six-digit alphanumeric representation of colors used in web design and digital graphi...

Hexadecimal (also known as hex or base-16) is a positional numeral system that uses 16 distinct symb...

HMAC (Hash-based Message Authentication Code) is a cryptographic technique that combines a secret ke...

HSL Color (Hue, Saturation, Lightness) is a cylindrical color model that represents colors in a way ...

HTML (HyperText Markup Language) is the standard markup language for creating web pages and web appl...

HTML Entity (HTML Character Entity) is a string that begins with an ampersand (&) and ends with a se...

HTTP (Hypertext Transfer Protocol) is the foundation protocol of the World Wide Web, defining how me...

HTTPS (Hypertext Transfer Protocol Secure) is the secure version of HTTP that encrypts all communica...

Human-in-the-Loop is a control pattern in which an AI system pauses at defined decision points so an...

Hybrid Search is a technique in information retrieval and RAG (Retrieval-Augmented Generation) syste...

HyDE is a retrieval technique that asks a language model to generate a hypothetical document or answ...

Image Compression is the process of reducing the file size of digital images by removing redundant o...

In-Context Learning (ICL) is the ability of large language models to learn and adapt to new tasks fr...

Indexer is a pipeline component that writes processed documents, chunks, embeddings, metadata, or sp...

Inference (model inference) is the execution of a trained model on new input to produce a prediction...

Instruction Tuning is a supervised fine-tuning approach that trains a pretrained model on tasks expr...

IP Address (Internet Protocol Address) is a unique numerical identifier assigned to every device con...

ISO 8601 is an international standard for representing dates and times in a clear, unambiguous forma...

Jailbreak is an adversarial attempt to make an AI model or model-backed system violate a specified s...

JavaScript is a high-level, interpreted programming language that enables interactive web pages and ...

JPEG (Joint Photographic Experts Group) is a commonly used lossy compression method for digital imag...

JSON (JavaScript Object Notation) is a lightweight, text-based data interchange format that is easy ...

JSON Mode is a provider-specific LLM output setting that makes a normally completed response syntact...

JSON Schema is a vocabulary that allows you to annotate and validate JSON documents. It describes th...

JSONPath is a query language for JSON documents that enables extracting and filtering data from comp...

JWT (JSON Web Token) is a compact, URL-safe means of representing claims to be transferred between t...

Knowledge Graph is a connected representation of entities, relationships, and claims, often with exp...

KTO is a preference tuning method that optimizes a language model using examples labeled as desirabl...

KV Cache (Key-Value Cache) is per-layer inference state that stores the attention Key and Value tens...

LangGraph is a low-level orchestration framework and runtime for building long-running, stateful wor...

Latency is the elapsed time between a request and a response or milestone in an AI system, such as f...

Lint is a static code analysis warning or error generated by a linter to flag style violations, susp...

Linting is the process of using automated tools to analyze source code for potential errors, bugs, s...

LLM (Large Language Model) is a neural language model trained to predict, score, or generate token s...

LLM-as-Judge is a probabilistic evaluation method in which a language model assesses one or more can...

LLMOps (Large Language Model Operations) is the engineering and governance discipline for developing...

LoRA (Low-Rank Adaptation) is a parameter-efficient adaptation method that keeps selected pretrained...

LoRA Rank is the inner dimension r of the two trainable matrices used to represent a LoRA weight upd...

Lorem Ipsum is a pseudo-Latin placeholder text commonly used in graphic design, web development, and...

Lost in the Middle is a position-sensitive long-context failure mode in which a language model uses ...

Machine Learning (ML) is a field of artificial intelligence in which an algorithm is fitted to data ...

Markdown is a lightweight markup language created by John Gruber in 2004 for formatting plain text d...

MCP (Model Context Protocol) is an open, versioned protocol introduced by Anthropic in 2024 for a Ho...

MCP App is an ecosystem label for a product that packages one or more Model Context Protocol capabil...

MCP Client is the protocol component instantiated by an MCP Host to communicate directly with one MC...

MCP Gateway is an optional, MCP-aware proxy or aggregation layer that routes Client requests to Serv...

MCP Host is the user-facing AI application that creates and coordinates MCP Clients, assembles model...

MCP Prompt is a user-controlled, server-authored message template that an MCP Client can discover an...

MCP Resource is an application-controlled, URI-addressed data contract that an MCP Server exposes fo...

MCP Sampling is a deprecated Model Context Protocol Client feature that lets a Server request model ...

MCP Server is the protocol endpoint that exposes declared tools, resources, and prompts to an MCP Cl...

MCP Tool is a model-controlled action contract that an MCP Server exposes for discovery and invocati...

MD5 (Message-Digest Algorithm 5) is a widely used cryptographic hash function that produces a 128-bi...

Metadata Filtering is the practice of restricting retrieval results using structured attributes atta...

Microsoft Agent Framework is Microsoft's open-source Python and .NET development kit for building AI...

MIME Type is a standardized identifier used to indicate the nature and format of a file or data, con...

Minification is the process of removing all unnecessary characters from source code without changing...

Mixture of Experts (MoE) is a conditional-computation architecture in which a learned or fixed routi...

Mock data is data deliberately constructed for development, testing, or demonstration to represent a...

Model Alignment is the ongoing engineering and research work of making an AI system pursue intended ...

Model Serving is the production practice of deploying machine learning or language models behind API...

Multi-Agent Systems are AI architectures in which multiple agents coordinate through explicit roles,...

Multimodal AI is an AI system that accepts, relates, or generates more than one data modality, such ...

NanoID is a tiny, secure, URL-friendly unique string ID generator for JavaScript. It generates compa...

Neural Network is a computational model inspired by the biological neural networks in the human brai...

NLP (Natural Language Processing) is a branch of artificial intelligence that focuses on enabling co...

OAuth (Open Authorization) is an open standard authorization protocol that allows users to grant thi...

Ollama is an open-source model runtime and distribution tool for running supported model artifacts o...

Open-Weight Model is an AI model whose trained parameters are made available for download under stat...

OpenTelemetry is an open-source observability framework that provides a unified set of APIs, SDKs, a...

ORPO is a preference optimization method that combines supervised learning on chosen responses with ...

Overfitting is a modeling error in machine learning that occurs when a model learns the training dat...

PagedAttention is an LLM serving technique that manages key-value cache memory in fixed-size blocks,...

Password is a secret string of characters used to authenticate a user's identity and grant access to...

PDF (Portable Document Format) is a file format developed by Adobe in 1992 to present documents cons...

PEFT (Parameter-Efficient Fine-Tuning) is a family of methods that adapts a pretrained model by trai...

Planner-Executor is an AI agent architecture that creates a versioned, machine-readable plan for a g...

PNG (Portable Network Graphics) is a raster graphics file format that supports lossless data compres...

PPO is a reinforcement learning algorithm that updates a policy while limiting how far each update m...

Preference Data is training data that records which model responses are preferred, ranked, rejected,...

Prefill is the LLM inference phase that processes the full input prompt in parallel and produces the...

Prettify is the process of formatting code or data to make it more readable by adding proper indenta...

Prompt is a natural language input or instruction given to an AI model to guide its response, servin...

Prompt CI/CD is the controlled integration, evaluation, release, monitoring, and rollback of changes...

Prompt Engineering is the practice of designing, refining, and optimizing input instructions (prompt...

Prompt Injection is attacker-controlled or attacker-influenced content that causes a large language ...

Prompt Regression Test is an evaluation that checks whether a prompt or related LLM application chan...

Prompt Template is a reusable prompt structure that combines fixed instructions with variable inputs...

Prompt Versioning is the practice of tracking prompt source history and binding each released LLM be...

QLoRA (Quantized Low-Rank Adaptation) is a parameter-efficient fine-tuning method that keeps a pretr...

QR Code (Quick Response Code) is a two-dimensional matrix barcode that can store various types of da...

Quantization is the process of mapping model weights, activations, or KV-cache values from a higher-...

Query Rewriting is the process of transforming a user's original question into one or more clearer, ...

Query String is the part of a URL that contains data to be passed to web applications, appearing aft...

RAG (Retrieval-Augmented Generation) is an AI architecture that enhances large language model output...

Rate limiting is a technique used to control the number of requests a client can make to an API or s...

Reciprocal Rank Fusion is a rank aggregation method that combines multiple retrieval result lists by...

Recommendation System is an information filtering system that predicts and suggests items or content...

Red Teaming is a structured, authorized adversarial assessment that attempts to cause harmful, insec...

Regex (Regular Expression) is a sequence of characters that defines a search pattern, primarily used...

Reinforcement Learning (RL) is a type of machine learning where an intelligent agent learns to make ...

Reranking is an advanced stage in information retrieval and RAG (Retrieval-Augmented Generation) pip...

REST (Representational State Transfer) is an architectural style for designing networked application...

Retriever is a query-to-context component that receives a user or agent query and returns relevant d...

Reward Model is a model trained to assign scores to candidate responses based on preference data or ...

RGB Color (Red Green Blue) is an additive color model where red, green, and blue light are combined ...

RGBA Color is an extension of the RGB color model that adds an alpha channel for opacity control, al...

RLHF (Reinforcement Learning from Human Feedback) is a family of training methods that use human jud...

RNN (Recurrent Neural Network) is a class of neural networks designed to process sequential data by ...

Router Agent is an AI agent or routing component that classifies a request and delegates it to the m...

RSA (Rivest-Shamir-Adleman) is an asymmetric cryptographic algorithm that uses a pair of keys - a pu...

Semantic Cache is a cache that embeds an incoming request, finds a sufficiently similar cached reque...

Semantic Search is information retrieval that uses learned representations to retrieve meaningfully ...

Sentiment analysis is a natural language processing (NLP) technique used to identify and extract sub...

Serialization is the process of converting complex data structures or objects into a format that can...

SFT (Supervised Fine-Tuning) is a fine-tuning method that continues training a pretrained model on l...

SHA-256 (Secure Hash Algorithm 256-bit) is a cryptographic hash function that produces a 256-bit (32...

Skill, in the Agent Skills sense, is a reusable directory centered on a SKILL.md file that gives a c...

Slug is a URL-friendly version of a string, typically derived from a title or name, that uses only l...

Small Language Model (SLM) is an informal, context-dependent category for a language model that has ...

Sparse Retrieval is a lexical search method that represents queries and documents with sparse term-w...

Speculative Decoding is an LLM inference technique where a faster draft model proposes multiple cand...

Speech Recognition, also called automatic speech recognition (ASR) or speech-to-text (STT), converts...

SQL (Structured Query Language) is a standardized programming language designed for managing and man...

SSE Transport is an HTTP-based Model Context Protocol transport pattern that uses Server-Sent Events...

SSL/TLS (Secure Sockets Layer / Transport Layer Security) are cryptographic protocols designed to pr...

State Space Model (SSM) represents a dynamic system or sequence with a latent state that is updated ...

STDIO Transport is a local Model Context Protocol transport that exchanges protocol messages between...

Structured Output is the practice of making an LLM return data in a predictable machine-readable for...

Supervised learning is a machine learning paradigm where models learn from labeled training data, us...

Supervisor Agent is the centralized control-loop role in a multi-agent system that owns shared task ...

SVG (Scalable Vector Graphics) is an XML-based vector image format for two-dimensional graphics that...

Synthetic Data is artificially generated data that mimics the statistical properties and patterns of...

System Prompt is the high-priority instruction layer that defines an LLM application's persistent ro...

Temperature is a hyperparameter in large language models that controls the randomness and creativity...

Tensor Parallelism is a model parallelism strategy that splits large neural network tensors and thei...

Test-Time Compute is a technique where AI models allocate additional computational resources during ...

Text-to-Image is a generative AI task that maps a natural-language prompt, and sometimes additional ...

Text-to-Video is a generative AI task that maps a natural-language prompt, and sometimes additional ...

Throughput is the amount of work a serving system completes per unit of time, such as requests per s...

A timestamp is a sequence of characters or encoded information identifying when a certain event occu...

Timezone is a region of the globe that observes a uniform standard time for legal, commercial, and s...

Token is the fundamental unit of text that Large Language Models (LLMs) process, representing a piec...

Tokenizer is the component that converts text into the token IDs a language model can process and de...

Tokens per Second is a throughput metric that measures how many output tokens an LLM generates per s...

TOML is a configuration file format designed to be easy to read due to its obvious semantics, mappin...

Tool Poisoning is an AI-agent attack in which malicious or compromised tool metadata, schemas, imple...

Tool Use is the system pattern in which a model selects or proposes an external capability, trusted ...

Training Data is the collection of labeled or unlabeled samples used to teach machine learning model...

Transformer is a neural-network architecture first described in the 2017 paper "Attention Is All You...

TTFT is the latency from sending an LLM request until the first generated token is received by the c...

TypeScript is a strongly typed programming language that builds on JavaScript by adding optional sta...

Unicode is a universal character encoding standard that assigns a unique number (code point) to ever...

Unit conversion is the process of expressing the same physical or measured quantity in a different u...

Unix Timestamp (also known as Unix Epoch Time or POSIX time) is a system for tracking time as a runn...

Unsupervised Learning is a machine learning paradigm where algorithms discover hidden patterns, stru...

URI is a string of characters that identifies a resource on the internet or within a system, providi...

URL is a specific type of URI that provides the complete address to locate a resource on the interne...

URL Encoding (Percent-Encoding) is a mechanism for encoding information in a Uniform Resource Identi...

UTF-8 (8-bit Unicode Transformation Format) is a variable-width character encoding that can represen...

UUID (Universally Unique Identifier) is a 128-bit identifier format standardized by RFC 9562. The ca...

VAE (Variational Autoencoder) is a generative deep learning model that combines neural network autoe...

A vector database is a specialized database designed to store, index, and query high-dimensional vec...

vLLM is an open-source library and server runtime for executing supported language, multimodal, embe...

VLM (Vision-Language Model) is a multimodal AI model that can jointly process and reason over both v...

WebLLM is an open-source, in-browser language-model inference engine from MLC AI. It loads compatibl...

WebP is a modern image format developed by Google that provides superior lossless and lossy compress...

Whitespace refers to characters that represent horizontal or vertical space in text but are typicall...

Workflow Orchestration is the structured coordination of tasks, dependencies, state, timers, signals...

XML (Extensible Markup Language) is a markup language and file format for storing, transmitting, and...

XPath is a query language for selecting nodes and computing values from XML documents using path exp...

YAML (YAML Ain't Markup Language) is a human-readable data serialization language commonly used for ...

Zero-Shot Learning is a family of transfer and evaluation settings in which a model predicts target ...