> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agents.labs.bandwidth.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How agents work

> Understand the choices that shape a voice conversation.

An agent combines instructions, a voice, and optional tools to complete a task through conversation.

1. **Define the task.** Choose one prompt or a workflow with several conversation steps.
2. **Choose AI connections.** Select the services used to understand speech, respond, and speak. A realtime connection combines speech input and output.
3. **Add context and tools.** Supply information about the caller and connect actions your agent can take.
4. **Test and publish.** Review a draft through a browser conversation before publishing it.
5. **Connect channels.** Use phone numbers, the browser tester, or the API to reach your agent.
6. **Review results.** Open Conversations to inspect transcripts, available recordings, extracted data, and tool results.

## Workspace defaults and agent settings

Your workspace's **AI connections** provide defaults. An agent can override selected connections for a particular language or use case. Changes to shared defaults can affect agents that inherit them, so test affected agents after changing a connection.

## Calls, runs, and campaigns

A **call** is a voice conversation. A **run** records an agent execution, including browser tests. A **campaign** organizes outbound calls to a list of contacts.

Learn about [agent types](/core-concepts/workflows-and-agents), [call results](/core-concepts/calls-and-runs), and [campaigns](/core-concepts/campaigns).