#ai-glossary

Chunking is the practice of splitting a larger piece of text or data into smaller pieces so a system can process, store, or retrieve it more effectively; semantic chunking does the same thing, but tries to cut at meaningful boundaries instead of by fixed length. Chunking is a core technique in retrieval-augmented generation (RAG), document search, summarization, and any workflow that needs to feed long content into an LLM. Why teams use it: LLMs have context limits, so very large documents must

Semantic search is a way to find information by meaning, not just by matching the exact words you typed. Traditional keyword search is good when you know the exact term to look for. Semantic search helps when users phrase the same idea in different words, use synonyms, or ask a question instead of naming a document. You’d reach for it when: search queries are messy or conversational, exact wording varies a lot across documents, you want to retrieve “about the same thing” rather than “contains th

An embedding is a vector embedding: a list of numbers that represents something—like a word, sentence, image, or user—in a way a model can compare mathematically. Embeddings turn messy, human data into a form machines can use for: Similarity search: “Find documents like this one.” Recommendations: “Show items close to what the user liked.” Clustering and classification: “Group related things together.” Retrieval for LLMs: “Fetch the most relevant context before generating an

A vector database, also called a vector store, is a system for storing and searching data by its numerical embeddings so you can find items that are “most similar” rather than exactly matching text or IDs. Regular databases are great when you know the exact thing you want: a user ID, a product SKU, a timestamp range. But many AI features need semantic search or nearest-neighbor retrieval: “find questions like this one,” “retrieve passages relevant to this prompt,” or “recommend items

Retrieval-augmented generation, or RAG, is a way to make a language model answer questions using relevant external documents it retrieves first, instead of relying only on what it memorized during training. RAG helps when you want answers that are: grounded in your own content, such as docs, policies, tickets, or product knowledge easier to update than retraining a model more likely to mention current or proprietary information In practice, teams reach for RAG when a model needs to answer “what

Tool use in AI agents is when an LLM-based agent can call external functions, APIs, or programs—often called a tool calling agent—to get information or take actions it cannot do reliably from text alone. A plain language model is good at generating text, but it does not reliably know live facts, cannot directly query a database, and should not guess at actions like sending email or creating tickets. Tool use solves that by letting the model delegate specific steps to purpose-built systems. Y

The agent loop, also called the perceive-plan-act loop, is the repeated cycle where an AI agent observes its situation, decides what to do next, and then takes an action. This is the basic control pattern behind many agentic systems. Instead of asking a model for one one-off answer, you let it keep checking the world, updating its plan, and acting until the task is done or it should stop. You reach for it when the task is dynamic, multi-step, or depends on intermediate results: calling tools, br

A ReAct agent is an LLM-driven system that alternates between reasoning and acting—often called reason + act or ReAct prompting—so it can decide what to do next, use tools, and update its plan from the results. ReAct is useful when a model should not just produce an answer, but also take steps in the world: search a document set, call an API, inspect a database, or ask for more information. In practice, teams reach for it when: a plain chat completion is too static, the task

An agentic workflow is a workflow where an AI system can decide and take multi-step actions toward a goal, instead of only producing a one-shot answer; you may also hear this called an agentic system. Use an agentic workflow when the task is not just “generate text,” but “figure out what to do next.” That includes tasks like investigating an issue, pulling together information from multiple tools, retrying after failure, or following conditional branches based on intermediate results

An AI agent, also called an LLM agent or autonomous agent, is a system that can pursue a goal by deciding what to do next, taking actions, and using the results to keep going. Most AI systems just answer a prompt once. An agent is useful when one answer is not enough and the system needs to act over multiple steps: search, call tools, inspect results, retry, and stop when the goal is met. You’d reach for an agent when the task is: open-ended, not a single fixed response dependent on external

An agent harness or agent runtime is the software layer that runs an AI agent, manages its loop of thinking and acting, and connects it to tools, memory, and external systems. An LLM by itself just returns text. An agent needs more than that: it has to decide when to call a tool, pass results back into the model, stop at the right time, handle errors, and keep track of state. That is what the harness/runtime does. It is the scaffolding around the model that turns “generate a response” in

Human-in-the-loop for agents means a person stays in the decision loop to review, approve, correct, or stop an AI agent before it takes important actions. Agentic systems can plan, call tools, and act across multiple steps. That is useful, but it also creates risk: a wrong tool call, a bad assumption, an unsafe side effect, or an action the user never intended. Human-in-the-loop adds a control point so the agent does not fully act on its own when the cost of a mistake is high. You reach for it w

A planner-executor agent is an AI system that first makes a plan, then carries out that plan step by step, usually by delegating the “thinking ahead” to one component and the “doing” to another. This pattern is useful when a task is too complex to solve in one shot. Instead of asking a model to answer, browse, call tools, and self-correct all at once, you split the work into two roles: Planner: decides the sequence of steps, subgoals, or tool calls. Executor: follows those steps and prod

Agent memory is the information an AI agent keeps so it can keep working across turns, and it usually means two different things: short-term memory for the current task context, and long-term memory for stored facts or past interactions it can retrieve later. Without memory, an agent is basically reset every time you ask it something new. That makes it bad at multi-step tasks, personalization, follow-up questions, and anything that needs continuity. In practice, memory helps an agent: re

A multi-agent system is a setup where multiple autonomous agents work separately or together to complete a task, often with some form of coordination or agent orchestration. A single agent is often enough for simple tasks, but multi-agent systems help when the problem is naturally split into parts: research, planning, execution, review, or parallel exploration. They can also let you assign different roles or tools to different agents, which is useful when one model instance would be too slow, to

An MCP server is a service that exposes tools, data, or actions to an AI client using the Model Context Protocol, so an LLM-powered app can safely discover and call external capabilities in a standard way. Without a standard protocol, every AI app ends up integrating with files, databases, APIs, and internal tools in its own custom way. That makes integrations harder to build, harder to reuse, and harder to govern. An MCP server solves that by giving AI clients a common interface for: discoverin

The Model Context Protocol (MCP) is an open standard that lets AI apps connect to external tools, data sources, and workflows through a common interface, usually via an MCP client talking to an MCP server. Without a shared protocol, every AI app tends to build one-off integrations for files, databases, APIs, and internal tools. MCP solves that by giving model-powered applications a standard way to discover and use capabilities from different systems. In practice, you reach for MCP when y

Function calling, also called tool calling or structured tool use, is a way for an AI model to ask your app to run a specific function with specific arguments instead of guessing the answer itself. Large language models are good at language, but they are not reliable sources of fresh facts, private data, or side effects like “send the email” or “look up today’s weather.” Function calling solves that by letting the model produce a structured request that your code can execute. You reach f

A tool schema or tool definition is a machine-readable description of a function, API, or action an LLM can call, including its name, what it does, and the inputs it accepts. Tool schemas let you connect a model to real capabilities without letting it invent the interface. In practice, they solve three common problems: Structured calling: the model can ask for a function call instead of writing brittle free-form text. Validation: the runtime can check that the model supplied the

A JSON schema for tool arguments is a machine-readable description of what inputs a tool expects, including the names, types, and constraints of its parameters. When an LLM calls a tool, it needs a reliable way to format the arguments so the tool can actually run. A JSON schema helps by telling the model and the application: which fields are required what type each field should be whether values must match a pattern, range, or enum how to validate the request before executing the tool In practic

Parallel tool calling is when an LLM asks for multiple tools to be run in the same turn, so independent actions can happen at the same time instead of one after another. Use parallel tool calling when the model needs several unrelated results, such as fetching weather for three cities, looking up multiple records, or calling a search API and a calculator in the same step. It matters because it can reduce latency and keep agent flows simpler: rather than forcing the model to serialize independent

A tool result or tool message is the message an app sends back to the model after the model asks it to use an external tool, such as a search API, calculator, database query, or function call. LLMs do not reliably know fresh data, private data, or exact computations on their own. A tool result lets the system return the tool’s output to the model in a structured way so the model can continue the conversation with real evidence. You’d use it whenever the model needs to: fetch data from an

A computer-use or browser-use tool is a capability that lets an AI agent operate software directly—usually a web browser, and sometimes the broader desktop—by clicking, typing, scrolling, and reading the visible screen instead of calling an API. Not every task has a clean API. Many real workflows still live in web apps: logging into dashboards, filling forms, checking inventories, downloading reports, or copying data between systems. A browser-use tool helps an LLM or agent handle those tasks by

Hybrid search is a retrieval method that combines keyword search with vector similarity search so you can find documents that match both the exact words a user typed and the broader meaning of the query. Pure keyword search is good when the exact term matters. Pure vector search is good when wording varies but meaning is similar. In practice, many real queries need both. You reach for hybrid search when: users may search with exact product names, error codes, or technical terms the same

A code execution tool, or sandbox, is a controlled runtime that lets an AI agent run code safely without giving it full access to your systems. LLMs are good at writing code, but they cannot reliably verify code by thought alone. A sandbox lets an agent: run generated code check outputs inspect errors iterate on fixes You’d reach for it when the task needs computation, file manipulation, data transformation, or testing. In practice, teams use sandboxes when “just answer in text” is too britt

Task decomposition is the practice of breaking a larger goal into smaller, manageable sub-tasks that can be solved one by one. Complex tasks are easier to plan, execute, and debug when they are split into parts. In AI systems, software projects, and agent workflows, decomposition helps you: reduce ambiguity in a broad request assign work to different components or steps catch errors earlier by checking intermediate results make progress on problems that are too large to solve in one shot In prac

Reflection, or self-critique, is when an AI agent reviews its own output or plan, spots mistakes or weaknesses, and then revises itself before acting. Agentic systems can make more than one kind of error: they may miss a constraint, choose a bad tool, hallucinate a fact, or produce a plan that looks plausible but won’t work. Reflection adds a built-in “check yourself” step, which is useful when you want better reliability than a single pass from the model. In practice, teams reach for reflection