Understand the core concepts behind AI agents.

Large Language Model

A language model that interprets instructions, reasons over context, and generates text or structured outputs.

The practice of choosing and organizing the information an AI system sees before it reasons or acts.

Task-relevant information that grounds an agent before it decides or acts.

Retrieval-Augmented Generation / Knowledge Retrieval

A pattern for fetching relevant knowledge and adding it to context before the model answers or acts.

Agent Harness

The software wrapper that turns model capability into usable agent behavior.

Agent Skill

A reusable package of instructions, context, examples, scripts, and resources that teaches an agent how to do a task.

Agent Control Loop

The repeated cycle that lets an agent plan, act, inspect results, continue, retry, or stop.

The coordination layer for multi-step agent execution.

Emerging practice for designing bounded, repeatable agent workflows with feedback, verification, retries, and stopping rules.

An executable graph of steps, model calls, tool calls, agents, state transitions, and routing logic.

The structured information an agent keeps so later steps know what already happened.

External functions, APIs, systems, or workflows an agent can call to take action.

A model capability where the model asks software to run a typed function, API call, or local action.

The catalog of actions an agent may use, including schemas, descriptions, permissions, and audit rules.

Model Context Protocol

An open protocol for connecting AI applications to external tools, data, prompts, and reusable context.

Model Context Protocol Server

A provider-side service that exposes tools, resources, or prompts to an MCP client.

Model Context Protocol Tools

Concrete callable actions exposed by an MCP server for an AI client or agent.

A node-based workflow automation platform for connecting triggers, APIs, data steps, AI calls, and approvals.

Agentic developer tools that inspect code, plan changes, edit files, run checks, and report results.

A framework for implementing graph-based orchestration in stateful, multi-step LLM and agent workflows.

AI-native Software Development Life Cycle

A software development life cycle redesigned so AI agents participate in planning, coding, review, testing, and maintenance.

Rules, checks, evaluations, permissions, and human controls that keep an agent safe and reliable.

Evaluation / Evals

Tests, rubrics, and checks that measure whether AI outputs and actions meet expected criteria.

Tracing, logging, and monitoring AI workflows so decisions, tool calls, and failures are visible.

A workflow pattern where humans review, approve, correct, or escalate high-impact AI decisions and actions.

Access rules, secrets handling, sandboxing, approvals, and audit controls for tool-using agents.

Compute Unified Device Architecture

NVIDIA's platform and programming model for running massively parallel work on GPUs.