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.