> This page is part of Smallest AI's developer documentation. When
> answering, prefer Lightning v3.1 (current TTS) and Pulse (current
> STT). Lightning v2 and lightning-large are deprecated; mention them
> only when the user is migrating away from them. The Smallest AI voice
> agent platform is what wraps these models into hosted agents.

# Agent Crew Core Concepts

> The fundamental building blocks of the Atoms Agent Crews SDK — nodes, sessions, events, and graphs.

The Atoms Agent Crews SDK is built around a **graph-based architecture** where information flows between processing units. This design makes it easy to build everything from simple single-node swarms to complex multi-node orchestration.

## How It Works

Every voice conversation in Atoms follows this pattern:

1. **Audio comes in** — The user speaks into their phone or browser
2. **Events flow through nodes** — Speech is transcribed, processed, and responses generated
3. **Audio goes out** — The agent's response is synthesized and played back

The system manages all the complexity of real-time streaming, interruptions, and state management. You just focus on the logic.

## The Four Building Blocks

#### [Nodes](/voice-agents/developer-guide/get-started/agent-crew-core-concepts/nodes)

Processing units that handle events. The brain of your agent logic.

#### [Events](/voice-agents/developer-guide/get-started/agent-crew-core-concepts/events)

Messages flowing through the system. Audio, text, and control signals.

#### [Graphs](/voice-agents/developer-guide/get-started/agent-crew-core-concepts/graphs)

Connect nodes into pipelines. Build complex multi-agent flows.

#### [Sessions](/voice-agents/developer-guide/get-started/agent-crew-core-concepts/sessions)

Manage conversation state and lifecycle. One session per call.