AI Agent Glossary

Core terms every agent builder should know. New to AI agents? Start with our complete beginner's guide or browse AI agent tools.

RAG

Retrieval-Augmented Generation: injects retrieved external context into model responses.

Tool Use

When an agent calls external tools/APIs to complete tasks.

Function Calling

Structured model output that triggers predefined code functions.

Chain of Thought

Intermediate reasoning process used to reach an answer.

ReAct

Pattern combining reasoning and tool actions in iterative loops.

Reflection

Self-critique step where an agent reviews and improves outputs.

Multi-Agent Systems

Multiple specialized agents coordinating on shared goals.

Agent Memory

Stored context that lets agents retain facts across steps or sessions.

Vector Embeddings

Numeric representations of meaning for semantic similarity search.

Semantic Search

Search by meaning, not exact keyword matching.

Chunking

Splitting documents into smaller units for better retrieval.

Prompt Engineering

Designing prompts and instructions to improve outcomes.

Fine-Tuning

Training a model on task-specific data to alter behavior.

Orchestration

Managing workflow, control flow, and agent/tool interactions.

Agent Loop

Repeated cycle of observe → reason → act → evaluate.

Planning

Breaking a goal into ordered sub-steps before execution.

Reasoning

Inference process used to decide actions and produce answers.

Grounding

Anchoring outputs to reliable data or trusted sources.

Hallucination

Confident but incorrect or fabricated model output.

Context Window

Maximum token length a model can process in one pass.

Token

A unit of text consumed and generated by language models.

Embedding Model

Model specialized for producing vector representations.

Retrieval

Fetching relevant content from memory, docs, or databases.

Agentic Workflow

Workflow where agents make decisions and take actions.

Human-in-the-Loop

Human approval or intervention at critical workflow points.

Autonomous Agent

Agent that can execute goals with minimal human input.

Semi-Autonomous

Agent that acts independently but with explicit constraints.

Agent Swarm

Large group of agents running in parallel with coordination rules.

Task Decomposition

Splitting complex work into smaller manageable tasks.

Agent Protocol

Standard for communication between agents and tools.

Tool Router

Logic for selecting the right tool for each action.

Guardrails

Rules and checks that constrain unsafe or invalid behavior.

Evaluation Harness

Systematic testing setup for quality and regression checks.

Observability

Tracing, metrics, and logs for understanding agent behavior.

Learn More About AI Agents