Free AI Agent Glossary Online - TestMu AI (Formerly LambdaTest)

AI Agent

Core

Software that uses AI to take autonomous actions on behalf of a user, including planning multi-step tasks and executing workflows.

LLM

Technical

Large 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

Technical

Retrieval-Augmented Generation. Retrieves relevant information from a knowledge base before generating a response, grounding the AI in factual data.

MCP

Technical

Model Context Protocol. An open protocol that standardizes how AI models connect to external tools and data sources.

Prompt Engineering

Core

Designing and refining prompts to get better, more consistent outputs from AI models.

Fine-Tuning

Technical

Further training a pre-trained AI model on specific data to improve performance on a particular use case.

Workflow Automation

Workflow

Using software to execute tasks automatically based on triggers and conditions. AI agents add intelligence by handling decisions.

No-Code / Low-Code

Tools

Platforms that allow building automations using visual interfaces instead of writing code.

API

Technical

Application Programming Interface. Rules that let software applications communicate. How AI agents connect to tools and data.

Hallucination

Core

When an AI generates plausible but factually incorrect information. RAG and grounding techniques reduce this risk.

Human-in-the-Loop

Workflow

A workflow where certain decisions require human approval before the AI proceeds.

Guardrails

Core

Constraints and rules on AI agents that prevent undesired behavior by limiting what agents can do, say, or access.

Token

Technical

Basic unit of text LLMs process, roughly 4 characters. Token counts determine API costs and context limits.

Context Window

Technical

Maximum text an AI model can process at once, measured in tokens. Larger windows handle longer documents and histories.

Vector Database

Technical

Specialized database that stores data as mathematical vectors, enabling AI to search by meaning rather than keywords.

Agentic Workflow

Workflow

A workflow where AI agents make decisions and choose tools autonomously rather than following a fixed path.

Multi-Agent System

Technical

Architecture where multiple specialized AI agents collaborate, each handling a specific domain.

Automation ROI

Business

Return on investment from automation: time saved and errors reduced minus tool and setup costs.

Lead Scoring

Business

Using AI to rank potential customers by their likelihood to buy based on behavioral and firmographic signals.

Churn Prediction

Business

Using AI to identify customers likely to cancel. Agents can act on churn signals with retention workflows.

Embedding

Technical

A numeric vector that represents the meaning of text, images, or other data so models can compare similarity.

Semantic Search

Technical

Search that matches meaning instead of exact keywords, usually powered by embeddings and a vector index.

Tool Calling

Technical

Letting an LLM request structured functions such as APIs, databases, or browsers so it can take real actions.

Function Calling

Technical

Provider term for tool use where the model emits structured arguments that a host application executes.

Temperature

Technical

Sampling setting that controls randomness. Lower values are more deterministic. Higher values are more varied.

Top-p

Technical

Also called nucleus sampling. Limits next-token choices to a probability mass, shaping how creative the output can be.

System Prompt

Core

Instructions that set role, tone, and constraints for the model before the user conversation starts.

Zero-Shot

Core

Asking a model to perform a task with no examples in the prompt, relying only on the instruction.

Few-Shot

Core

Providing a small number of examples in the prompt so the model can follow the pattern.

Chain of Thought

Core

Prompting style that asks the model to reason step by step before giving a final answer.

ReAct

Workflow

Reasoning and acting loop where an agent alternates between thinking, calling tools, and observing results.

Orchestration

Workflow

Coordinating models, tools, memory, and steps so a multi-step agent workflow runs in the right order.

Subagent

Technical

A specialized agent spawned by a parent agent to handle one focused subtask and return a result.

Memory

Technical

Stored state an agent can recall across turns or sessions, such as chat history or long-term facts.

Episodic Memory

Technical

Memory of specific past interactions or events an agent can retrieve when facing a similar situation.

Grounding

Core

Any technique that anchors model output in verifiable facts, documents, or tools instead of free invention.

Knowledge Cutoff

Core

The date after which a model has no trained knowledge unless you supply newer information at runtime.

Prompt Injection

Testing

Attack where untrusted text tries to override system instructions and steer the model toward unsafe actions.

Jailbreak

Testing

Prompt pattern that attempts to bypass safety policies so the model produces restricted content.

Alignment

Core

Training and policy work that steers a model toward helpful, honest, and harmless behavior.

RLHF

Technical

Reinforcement Learning from Human Feedback. Fine-tunes a model using ranked human preferences.

LoRA

Technical

Low-Rank Adaptation. Efficient fine-tuning method that updates small adapter weights instead of the full model.

Quantization

Technical

Reducing model weight precision to shrink size and cost while keeping most of the original quality.

Distillation

Technical

Training a smaller model to imitate a larger one so inference is cheaper and faster.

Latency

Business

Time from request to first or final response. Critical for chat UX and agent step loops.

Throughput

Business

How many requests or tokens a system can process in a given time window under load.

Token Cost

Business

Price billed for prompt and completion tokens. Directly drives the cost of agent workflows.

Structured Output

Technical

Forcing the model to return JSON or another schema so downstream code can parse results reliably.

JSON Mode

Tools

API setting that asks the model to respond with valid JSON for safer parsing in applications.

Streaming

Tools

Sending tokens to the client as they are generated so users see partial answers sooner.

Chunking

Technical

Splitting documents into smaller pieces before embedding so retrieval stays precise and affordable.

Reranking

Technical

Second-pass scoring of retrieved chunks so the most relevant passages reach the model first.

Hybrid Search

Technical

Combining keyword search and vector search to catch both exact matches and semantic matches.

Agent Loop

Workflow

Repeat cycle of gather context, take action, and verify results until the goal is met or stopped.

Planner

Workflow

Component that breaks a goal into ordered steps before tools are called.

Executor

Workflow

Component that runs planned steps, calls tools, and returns observations to the agent.

Supervisor Agent

Workflow

Agent that routes work to specialist agents and merges their outputs into a final result.

Skill

Tools

Reusable instruction pack that teaches an agent how to perform a recurring task in a project.

AGENTS.md

Tools

Project file that gives coding agents repo-specific rules, conventions, and workflow guidance.

MCP Server

Tools

Process that exposes tools, resources, and prompts over the Model Context Protocol.

MCP Client

Tools

Connector inside a host app that talks to MCP servers on behalf of the model.

Sandbox

Testing

Isolated environment where agent code or tool calls run with limited network and file access.

Eval

Testing

Systematic test of model or agent quality using scored tasks, rubrics, or golden datasets.

Golden Dataset

Testing

Curated set of inputs and expected outputs used to measure regressions in agent behavior.

Regression Test

Testing

Repeatable check that a prompt, tool, or agent still passes after a model or code change.

Observability

Testing

Tracing prompts, tool calls, latency, and errors so teams can debug agent runs in production.

Trace

Testing

Record of one agent run showing steps, tool inputs, outputs, and timing for later review.

Red Teaming

Testing

Adversarial testing that probes agents for prompt injection, data leaks, and policy failures.

Autonomy Level

Business

How much an agent can decide and act without asking a person. Higher levels need stronger guardrails.

SLA for Agents

Business

Service targets for agent workflows such as response time, success rate, and human escalation windows.

Copilot

Business

AI assistant that suggests actions while a person stays in control of final decisions.

Autonomous Agent

Core

Agent designed to pursue goals with minimal supervision, choosing tools and next steps on its own.

Knowledge Base

Tools

Document store an agent can query at runtime, often through RAG, to answer with private or fresh facts.

Citation

Core

Link or reference back to the source chunk used in an answer so users can verify claims.

Fallback

Workflow

Backup path when a tool fails or confidence is low, such as retrying, asking a human, or returning a safe default.

Rate Limit

Tools

Cap on how many API calls a client can make in a period. Agents must respect it or queue work.

Idempotency

Workflow

Design so repeating the same agent action does not create duplicate side effects like double charges.

Determinism

Testing

How consistently a model returns the same answer for the same input. Useful when writing stable tests.

Flaky Agent Test

Testing

Test that passes and fails intermittently because of model randomness, timing, or external tool variance.

Prompt Template

Tools

Reusable prompt with placeholders for variables so teams keep wording consistent across runs.

Persona

Core

Role and style instructions that shape how an agent speaks and what it prioritizes.

Multimodal

Technical

Models or agents that accept more than one input type, such as text plus images or audio.

Vision Model

Technical

Model that can interpret images or screenshots, useful for UI agents and document extraction.

Speech-to-Text

Tools

Converts spoken audio into text that an agent can reason over or act on.

Text-to-Speech

Tools

Converts agent text replies into spoken audio for voice interfaces.

Browser Agent

Tools

Agent that can navigate web pages, click, type, and extract content to complete tasks.

Coding Agent

Tools

Agent specialized in reading repos, editing code, running tests, and opening pull requests.

Test Generation Agent

Testing

Agent that drafts unit, API, or UI tests from requirements, code, or recorded sessions.

Self-Healing Locator

Testing

Automation pattern that repairs broken UI selectors when the page structure changes.

Synthetic Data

Testing

Artificially generated inputs used to expand coverage when real production data is scarce or sensitive.

Canary Prompt

Testing

Fixed probe prompt run after each model update to catch sudden quality drops early.

Cost Guardrail

Business

Budget limit or stop condition that prevents an agent loop from spending unbounded tokens.

Escalation Path

Business

Defined handoff from an agent to a human when confidence is low or policy requires approval.

Vendor Lock-in

Business

Risk of depending on one model or agent platform so deeply that switching later becomes costly.

Model Router

Tools

Layer that picks which model handles a request based on cost, latency, or task type.

Context Engineering

Core

Craft of choosing what history, tools, and documents enter the prompt so the model stays focused.

Prompt Caching

Technical

Reusing a stable prompt prefix across calls to cut latency and token cost on long system instructions.

Stop Sequence

Technical

Token pattern that tells the model to end generation early, useful for structured formats.

Max Tokens

Technical

Hard cap on how many tokens a model may generate in one response.

Tool Schema

Tools

JSON description of a tool name, parameters, and types that the model must follow when calling it.