> 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.

# Voice Agents

> What the Smallest AI voice agent platform is, the three ways to build an agent (dashboard, Agent Crew SDK, API), and where to go next.

A voice agent on Smallest is a configured agent (prompt, voice, language, tools, knowledge base) plus a way to reach it: a phone number, a web widget, or a mobile SDK. Calls run on Pulse for speech to text and Lightning for text to speech, with the LLM turn either handled by the platform or by your own code.

## Three ways to build

#### [Dashboard](/voice-agents/quickstart)

![](/_fern-img/624c2e4568979cf9b4780a0096a4fd86faa782cec18f37c9bee232a1e7e5dd7f.webp)

Pick a template or start from scratch, write the prompt, choose a voice, test in the browser. Five minutes to a working agent.

#### [Agent Crew SDK](/voice-agents/developer-guide/get-started/quickstart)

![](/_fern-img/af2186b5bbcbaf6fc63ddb945fde59562c85f533ea8d005f0595c05bf96f9244.webp)

Your code owns the LLM turn over a realtime WebSocket. Smallest owns speech, telephony, and turn taking.

#### [API](/api-reference/voice-agents)

![](/_fern-img/a0873498183a38b253e7a583d2b53e21fc5a301ab4b445acb1dad20d32e900fd.webp)

Create agents, place calls, run campaigns, and read analytics from your own backend.

Most teams start in the dashboard, then move to the API or SDK once the agent design settles. Everything configured in the dashboard is a JSON attribute on the agent, so the same guide applies to every path.

## What you can do

#### [Build](/voice-agents/build/build-your-agent)

Prompt, speech settings, variables, knowledge base, reusable tools, prompt scoring, versioning.

#### [Telephony](/voice-agents/telephony/phone-numbers)

Rent a number, bring your own SIP trunk, build audiences, run outbound campaigns.

#### [Integrate](/voice-agents/integrate/embed-a-voice-agent)

Web widget, Agent Web SDK, React Native, iOS, Android, Flutter, webhooks.

#### [Operate](/voice-agents/operate/analytics)

Analytics, evaluations, conversation logs, post call metrics.

## For AI coding agents

Install the [MCP server](/voice-agents/mcp/getting-started/quick-start) and ask Claude Code, Cursor, or Codex to create, configure, and call an agent in plain English. The [Build with a coding agent](/overview/developer-tools/build-with-a-coding-agent) page has a paste ready prompt that walks a coding agent from discovery to a live test call.

## Next steps

* [Quickstart](/voice-agents/quickstart): create and test an agent in the dashboard.
* [Build your agent](/voice-agents/build/build-your-agent): every field on an agent, in the order to fill them in.
* [Phone Numbers](/voice-agents/telephony/phone-numbers): put the agent on a real number.
* [Glossary](/voice-agents/resources/glossary): the terms used across the platform.

> Hosted voice agents for phone, web, and mobile. Smallest runs speech to text, the LLM turn, text to speech, and telephony. You bring the prompt, the tools, and the business logic.