Moon

The fastest, ultra realistic
voice AI platform

Powered by high performance State Space Model technology.

Start for free Read the docs
Welcome to Moon. It is great to meet you.
Make a phone call
Read this article
Narrate a scene
🇺🇸 Sophie ▾

A voice that reads back
in the color of the moment.

Moon renders a hundred voices, forty languages, and every emotional register your product needs, at latencies under two hundred milliseconds.

SophieUS English, warm register
There is a particular kind of quiet that follows a well made sentence. Moon tries, every time, to leave that quiet in the room.
18 seconds, streamed142ms first byte

A hundred voices, forty languages.

Cast the exact narrator, agent or character your product needs. Warm, sharp, whispered, breathless. Every voice ships with the range you would expect from a real one.

Real time conversation, not just playback.

Moon turns your voice into a real time agent that listens, responds, and interrupts naturally. Sub two hundred millisecond first byte, no compromises on quality.

An SDK your engineers
keep quietly happy.

One import, one line to speak, one line to stream. Every voice ships with typed schemas and predictable latency. Drop it into a call center or a game engine the same afternoon.

Sena
Japanese, whispered
Marcus
UK English, warm
Rae
US English, playful
Priya
Hindi, editorial
Tomás
Spanish, cinematic
Iris
French, close mic
Hey Sophie, can you check my order and tell me when it lands?
Caller, 0.9 seconds
Let me pull it up. Your two piece order ships this afternoon and arrives Thursday, likely before noon. Anything else I can move for you?
Moon agent, first byte 142ms
Actually can you push it to Friday? I am traveling.
Caller, 1.4 seconds
Done. Friday delivery, before 3pm, held at the front desk. I sent a note to your inbox too.
Moon agent, task committed
agent.ts
// Cast a voice, stream a response

import { Moon } from "@moon/client"



const voice = new Moon({ cast: "sophie" })



for await (const chunk of voice.speak(text)) {

  audio.write(chunk)

}

// First byte in 142ms, streamed to the ear.

Three quiet things Moon gets right.

The voice, not the vocoder

Moon renders breath, hesitation, and the tiny half smile inside a word. Every voice ships with the range you would expect from a real one.

Latency you can build on

Sub two hundred millisecond first byte, region to region. Real time enough for a phone call, predictable enough for a soak test.

An SDK that reads well

One import, one line to speak. Typed schemas, streaming primitives, and the specific control a production engineer expects.

Moon is the first voice API I have shipped in production
without asking a user to forgive the machine.

Marcus BeltránHead of Product, Northline

Six lines of code.
A voice your users will remember.

Drop in the Moon SDK, cast a voice, stream to the ear. The whole conversation lives in a single import.

app/voice.ts
// Cast a voice, stream a response, ship the same day

import { Moon } from "@moon/client"



const voice = new Moon({ cast: "sophie", warmth: "close" })



for await (const chunk of voice.speak(text)) {

  audio.write(chunk)

}

// First byte in 142ms, streamed to the ear.
Moon SDK 1.4In production, four hundred teams

Ship a voice
this afternoon.

Free for the first hundred thousand seconds. Bring a script, we bring the voice, and the first byte arrives in an hour.