Skip to navigation

Smallest AI Documentation

Build voice agents, or call the speech and language models directly. Everything you need is in one place.
View as Markdown

Choose your path

Voice Agents

Agents that keep the conversation going. Any language. Any channel. Dashboard, SDK, and API.

Build an agent
Models

Text to speech, speech to text, speech to speech, and an LLM. HTTP and WebSocket from any stack.

Call the models
API Reference

Every endpoint for both APIs, with schemas and copyable snippets in cURL, Python, and TypeScript.

Browse endpoints

How Smallest AI works

Smallest AI is voice infrastructure. Two product lines share one account, one API key, and one set of docs.

  • Voice Agents is the hosted platform for agents that talk to your customers over the phone, on the web, or inside a mobile app. It runs speech to text, the LLM turn, text to speech, and telephony for you. Configure an agent in the dashboard, or write the LLM turn yourself with the Agent Crew SDK.
  • Models are the individual engines behind it: Lightning (text to speech), Pulse (speech to text), Hydra (speech to speech), and Electron (LLM). Call them over HTTP or WebSocket from any stack, including Pipecat, LiveKit, and the Vercel AI SDK.

Meet the models

Seven models, four jobs. Compare them on one page, or open a card.

Lightning v3.1 Pro
Text to speech

Premium text to speech. Curated voices across American, British, and Indian accents, English and Hindi with code switching, plus 27 more languages.

Lightning v3.1
Text to speech

Low latency text to speech at 44.1 kHz. 12 languages and instant voice cloning from a short sample.

Pulse 2.0
Speech to text

English-only streaming speech to text with built-in end-of-turn detection, per-utterance emotion, and gender.

Pulse
Speech to text

Multilingual speech to text for realtime and pre-recorded audio. Diarization, redaction, keyword boosting, timestamps.

Pulse Pro
Speech to text

English speech to text tuned for accuracy. Pre-recorded audio only.

Electron
LLM

OpenAI compatible LLM built for voice agents. Fast first token, 70 languages, tool calling, prefix caching.

Hydra
Speech to speech

Full duplex speech to speech over one WebSocket. Built in barge-in and tool calling. English today.

See the Models overview for the full lineup and how to select each model.

Browse by capability

Text to Speech

Turn text into speech over HTTP, SSE, or WebSocket.

Lightning v3.1 ProLightning v3.1
Speech to Text

Transcribe files or live audio streams.

Pulse 2.0PulsePulse Pro
Speech to Speech

Audio in, audio out, with no pipeline to assemble.

Hydra
LLM

Chat completions with a drop in OpenAI compatible API.

Electron
Voice Cloning

Clone a voice from a few seconds of audio.

Telephony

Rent numbers, connect SIP trunks, run outbound campaigns.

Web and mobile

Embed an agent in a web page or a React Native, iOS, Android, or Flutter app.

Integrations

LiveKit, Pipecat, Vercel AI SDK, n8n, telephony providers, and more.

Self Host

Run the models in your own cloud with Docker or Kubernetes.

Building with an AI coding agent

The docs are written to be read by Claude Code, Cursor, Codex, and similar tools as much as by people.

  • /llms.txt is the machine readable index. Append .md to any page URL for clean Markdown.
  • The MCP server lets a coding agent create and configure agents in plain English.
  • Build with a coding agent has paste ready prompts for the platform and for the models.

Get help

Join the Discord or email support@smallest.ai. Check status.smallest.ai for service status across every product, and subscribe there for incident notifications. Account, billing, and team settings are under Administration.