World’s largest virtual agentic engineering & quality conference
This free tool allows you to search plain-English definitions for AI agent, LLM, RAG, and automation terms.
Type a word to filter by name or definition. Nothing you type leaves your browser.
Choose one topic filter, or keep All selected to see every matching term.
100 terms found
AI Agent
CoreSoftware that uses AI to take autonomous actions on behalf of a user, including planning multi-step tasks and executing workflows.
LLM
TechnicalLarge Language Model. A neural network trained on vast text data that can understand and generate human-like language. Examples include GPT-4, Claude, and Gemini.
RAG
TechnicalRetrieval-Augmented Generation. Retrieves relevant information from a knowledge base before generating a response, grounding the AI in factual data.
MCP
TechnicalModel Context Protocol. An open protocol that standardizes how AI models connect to external tools and data sources.
Prompt Engineering
CoreDesigning and refining prompts to get better, more consistent outputs from AI models.
Fine-Tuning
TechnicalFurther training a pre-trained AI model on specific data to improve performance on a particular use case.
Workflow Automation
WorkflowUsing software to execute tasks automatically based on triggers and conditions. AI agents add intelligence by handling decisions.
No-Code / Low-Code
ToolsPlatforms that allow building automations using visual interfaces instead of writing code.
API
TechnicalApplication Programming Interface. Rules that let software applications communicate. How AI agents connect to tools and data.
Hallucination
CoreWhen an AI generates plausible but factually incorrect information. RAG and grounding techniques reduce this risk.
Human-in-the-Loop
WorkflowA workflow where certain decisions require human approval before the AI proceeds.
Guardrails
CoreConstraints and rules on AI agents that prevent undesired behavior by limiting what agents can do, say, or access.
Token
TechnicalBasic unit of text LLMs process, roughly 4 characters. Token counts determine API costs and context limits.
Context Window
TechnicalMaximum text an AI model can process at once, measured in tokens. Larger windows handle longer documents and histories.
Vector Database
TechnicalSpecialized database that stores data as mathematical vectors, enabling AI to search by meaning rather than keywords.
Agentic Workflow
WorkflowA workflow where AI agents make decisions and choose tools autonomously rather than following a fixed path.
Multi-Agent System
TechnicalArchitecture where multiple specialized AI agents collaborate, each handling a specific domain.
Automation ROI
BusinessReturn on investment from automation: time saved and errors reduced minus tool and setup costs.
Lead Scoring
BusinessUsing AI to rank potential customers by their likelihood to buy based on behavioral and firmographic signals.
Churn Prediction
BusinessUsing AI to identify customers likely to cancel. Agents can act on churn signals with retention workflows.
Embedding
TechnicalA numeric vector that represents the meaning of text, images, or other data so models can compare similarity.
Semantic Search
TechnicalSearch that matches meaning instead of exact keywords, usually powered by embeddings and a vector index.
Tool Calling
TechnicalLetting an LLM request structured functions such as APIs, databases, or browsers so it can take real actions.
Function Calling
TechnicalProvider term for tool use where the model emits structured arguments that a host application executes.
Temperature
TechnicalSampling setting that controls randomness. Lower values are more deterministic. Higher values are more varied.
Top-p
TechnicalAlso called nucleus sampling. Limits next-token choices to a probability mass, shaping how creative the output can be.
System Prompt
CoreInstructions that set role, tone, and constraints for the model before the user conversation starts.
Zero-Shot
CoreAsking a model to perform a task with no examples in the prompt, relying only on the instruction.
Few-Shot
CoreProviding a small number of examples in the prompt so the model can follow the pattern.
Chain of Thought
CorePrompting style that asks the model to reason step by step before giving a final answer.
ReAct
WorkflowReasoning and acting loop where an agent alternates between thinking, calling tools, and observing results.
Orchestration
WorkflowCoordinating models, tools, memory, and steps so a multi-step agent workflow runs in the right order.
Subagent
TechnicalA specialized agent spawned by a parent agent to handle one focused subtask and return a result.
Memory
TechnicalStored state an agent can recall across turns or sessions, such as chat history or long-term facts.
Episodic Memory
TechnicalMemory of specific past interactions or events an agent can retrieve when facing a similar situation.
Grounding
CoreAny technique that anchors model output in verifiable facts, documents, or tools instead of free invention.
Knowledge Cutoff
CoreThe date after which a model has no trained knowledge unless you supply newer information at runtime.
Prompt Injection
TestingAttack where untrusted text tries to override system instructions and steer the model toward unsafe actions.
Jailbreak
TestingPrompt pattern that attempts to bypass safety policies so the model produces restricted content.
Alignment
CoreTraining and policy work that steers a model toward helpful, honest, and harmless behavior.
RLHF
TechnicalReinforcement Learning from Human Feedback. Fine-tunes a model using ranked human preferences.
LoRA
TechnicalLow-Rank Adaptation. Efficient fine-tuning method that updates small adapter weights instead of the full model.
Quantization
TechnicalReducing model weight precision to shrink size and cost while keeping most of the original quality.
Distillation
TechnicalTraining a smaller model to imitate a larger one so inference is cheaper and faster.
Latency
BusinessTime from request to first or final response. Critical for chat UX and agent step loops.
Throughput
BusinessHow many requests or tokens a system can process in a given time window under load.
Token Cost
BusinessPrice billed for prompt and completion tokens. Directly drives the cost of agent workflows.
Structured Output
TechnicalForcing the model to return JSON or another schema so downstream code can parse results reliably.
JSON Mode
ToolsAPI setting that asks the model to respond with valid JSON for safer parsing in applications.
Streaming
ToolsSending tokens to the client as they are generated so users see partial answers sooner.
Chunking
TechnicalSplitting documents into smaller pieces before embedding so retrieval stays precise and affordable.
Reranking
TechnicalSecond-pass scoring of retrieved chunks so the most relevant passages reach the model first.
Hybrid Search
TechnicalCombining keyword search and vector search to catch both exact matches and semantic matches.
Agent Loop
WorkflowRepeat cycle of gather context, take action, and verify results until the goal is met or stopped.
Planner
WorkflowComponent that breaks a goal into ordered steps before tools are called.
Executor
WorkflowComponent that runs planned steps, calls tools, and returns observations to the agent.
Supervisor Agent
WorkflowAgent that routes work to specialist agents and merges their outputs into a final result.
Skill
ToolsReusable instruction pack that teaches an agent how to perform a recurring task in a project.
AGENTS.md
ToolsProject file that gives coding agents repo-specific rules, conventions, and workflow guidance.
MCP Server
ToolsProcess that exposes tools, resources, and prompts over the Model Context Protocol.
MCP Client
ToolsConnector inside a host app that talks to MCP servers on behalf of the model.
Sandbox
TestingIsolated environment where agent code or tool calls run with limited network and file access.
Eval
TestingSystematic test of model or agent quality using scored tasks, rubrics, or golden datasets.
Golden Dataset
TestingCurated set of inputs and expected outputs used to measure regressions in agent behavior.
Regression Test
TestingRepeatable check that a prompt, tool, or agent still passes after a model or code change.
Observability
TestingTracing prompts, tool calls, latency, and errors so teams can debug agent runs in production.
Trace
TestingRecord of one agent run showing steps, tool inputs, outputs, and timing for later review.
Red Teaming
TestingAdversarial testing that probes agents for prompt injection, data leaks, and policy failures.
Autonomy Level
BusinessHow much an agent can decide and act without asking a person. Higher levels need stronger guardrails.
SLA for Agents
BusinessService targets for agent workflows such as response time, success rate, and human escalation windows.
Copilot
BusinessAI assistant that suggests actions while a person stays in control of final decisions.
Autonomous Agent
CoreAgent designed to pursue goals with minimal supervision, choosing tools and next steps on its own.
Knowledge Base
ToolsDocument store an agent can query at runtime, often through RAG, to answer with private or fresh facts.
Citation
CoreLink or reference back to the source chunk used in an answer so users can verify claims.
Fallback
WorkflowBackup path when a tool fails or confidence is low, such as retrying, asking a human, or returning a safe default.
Rate Limit
ToolsCap on how many API calls a client can make in a period. Agents must respect it or queue work.
Idempotency
WorkflowDesign so repeating the same agent action does not create duplicate side effects like double charges.
Determinism
TestingHow consistently a model returns the same answer for the same input. Useful when writing stable tests.
Flaky Agent Test
TestingTest that passes and fails intermittently because of model randomness, timing, or external tool variance.
Prompt Template
ToolsReusable prompt with placeholders for variables so teams keep wording consistent across runs.
Persona
CoreRole and style instructions that shape how an agent speaks and what it prioritizes.
Multimodal
TechnicalModels or agents that accept more than one input type, such as text plus images or audio.
Vision Model
TechnicalModel that can interpret images or screenshots, useful for UI agents and document extraction.
Speech-to-Text
ToolsConverts spoken audio into text that an agent can reason over or act on.
Text-to-Speech
ToolsConverts agent text replies into spoken audio for voice interfaces.
Browser Agent
ToolsAgent that can navigate web pages, click, type, and extract content to complete tasks.
Coding Agent
ToolsAgent specialized in reading repos, editing code, running tests, and opening pull requests.
Test Generation Agent
TestingAgent that drafts unit, API, or UI tests from requirements, code, or recorded sessions.
Self-Healing Locator
TestingAutomation pattern that repairs broken UI selectors when the page structure changes.
Synthetic Data
TestingArtificially generated inputs used to expand coverage when real production data is scarce or sensitive.
Canary Prompt
TestingFixed probe prompt run after each model update to catch sudden quality drops early.
Cost Guardrail
BusinessBudget limit or stop condition that prevents an agent loop from spending unbounded tokens.
Escalation Path
BusinessDefined handoff from an agent to a human when confidence is low or policy requires approval.
Vendor Lock-in
BusinessRisk of depending on one model or agent platform so deeply that switching later becomes costly.
Model Router
ToolsLayer that picks which model handles a request based on cost, latency, or task type.
Context Engineering
CoreCraft of choosing what history, tools, and documents enter the prompt so the model stays focused.
Prompt Caching
TechnicalReusing a stable prompt prefix across calls to cut latency and token cost on long system instructions.
Stop Sequence
TechnicalToken pattern that tells the model to end generation early, useful for structured formats.
Max Tokens
TechnicalHard cap on how many tokens a model may generate in one response.
Tool Schema
ToolsJSON description of a tool name, parameters, and types that the model must follow when calling it.
An AI agent glossary is a searchable reference that explains agent and large language model vocabulary in plain English. You type a term such as RAG, MCP, hallucination, or guardrails and get a short definition you can use in docs, reviews, and vendor calls. This TestMu AI tool keeps the full list in your browser so lookups stay private and fast.
Agent products ship new jargon every quarter. Teams lose time arguing over names instead of architecture. A shared glossary fixes that. Product, QA, and engineering can point to the same meaning for token, context window, eval, or human-in-the-loop before a design review starts.
Misread terms create expensive mistakes. If you treat a chatbot like an autonomous agent, you underbuild guardrails. If you confuse grounding with RAG, you buy the wrong retrieval stack. Clear definitions keep reviews concrete.
Lookups take a few seconds. Keep the page open beside a model card, RFC, or PR and jump to any term you do not recognize.
The widget is built for quick reference during real work, not for scrolling a static blog post. Here is what you get:
Use the glossary whenever agent vocabulary shows up mid-task and you need a crisp meaning before you continue.
TestMu AI maintains this glossary beside its broader free online tools set, the same team behind cross-browser and real-device testing across 10,000+ real devices and 3000+ browsers. All processing happens in your browser. No data is uploaded.
An AI agent glossary is a searchable reference of plain-English definitions for agent and LLM terms. You look up words like RAG, MCP, token, and guardrails without reading a full textbook. This free TestMu AI tool keeps every definition in your browser.
Type into the search box. Matches cover both the term name and the definition text, so a query like vector surfaces embeddings and semantic search too. Category chips further narrow Core, Technical, Business, Tools, Workflow, or Testing entries.
No. Filtering runs entirely in your browser with a local term list. Nothing you type is sent to a server, stored in cookies, or used to train a model. Close the tab and the query is gone.
A chatbot mainly generates replies in conversation. An AI agent also plans steps, calls tools, and can change external systems. Chatbots talk. Agents act, which is why guardrails and human approval matter more for agent workflows.
RAG means Retrieval-Augmented Generation. The system fetches relevant documents first, then asks the model to answer using that context. It reduces hallucinations on private or fresh facts that were never in the model weights.
Product managers, QA engineers, developers, founders, and operators who keep hitting terms like MCP, LoRA, or human-in-the-loop in docs and vendor decks. It is built for quick lookups while you build or review agent systems.
Yes. The Testing category includes eval, golden dataset, prompt injection, red teaming, flaky agent tests, and related QA language. That focus helps teams ship agents with the same discipline they already use for software quality.
Yes. The tool is free with no signup, no usage cap, and no paywall on definitions. Search, filter, and copy as often as you need. TestMu AI maintains it as part of the free online tools collection.
Did you find this page helpful?
TestMu AI forEnterprise
Get access to solutions built on Enterprise
grade security, privacy, & compliance