Glossary

Agent

An agent is a system that pursues a goal by choosing actions, using tools, observing results, and iterating until completion or halt.

Agent Scaffolding

Scaffolding is the software around a model—loops, tools, memory, and policies—that turns completions into goal-seeking agent behavior.

Context Window

The context window is the maximum amount of tokens a model can consider at once, including instructions, history, tool results, and outputs.

Function Calling

Function calling is a model capability to emit structured tool invocations (name + arguments) that a runtime executes and returns as observations.

Guardrail

A guardrail is a control that constrains agent behavior—permissions, filters, budgets, or approval gates—enforced outside free-form model text.

MCP (Model Context Protocol)

MCP is an open protocol for connecting AI applications to external tools and data sources through standardized servers.

Memory

Agent memory is retained state across steps or sessions, including working context, thread history, durable facts, and retrieved knowledge.

Multi-Agent System

A multi-agent system coordinates multiple specialized agents that collaborate via messages, supervisors, or shared state.

Orchestrator

An orchestrator is the runtime that coordinates agent steps, tools, state transitions, and optionally multiple agents.

Planner

A planner is the component (often a model prompt or node) that decomposes a goal into steps before or during execution.

Tool

A tool is an external capability an agent can invoke—such as an API, database query, browser, or code runner—via a structured interface.

Trajectory

A trajectory is the ordered record of an agent run: thoughts or decisions, tool calls, observations, and outcomes.