JAJarvis
Dry wit, composed British AI assistant
en
The open-source AI voice studio
Clone, dictate and create.
Clone voices, generate speech across seven TTS engines, dictate into any app, and talk to agents in voices you own. A local-first alternative to cloud voice tools, running entirely on your machine.
Voices
Six profiles ready for the studio.
MOMorgan
Warm editorial narrator with a measured pace
en
FIFireship
Rapid-fire, deadpan tech humor with zero filler
en
SCScarlett
Smooth low alto with understated warmth
en
PRPrime
The name is the Primeagen
en
BEBen
A pixel-perfect kinda guy
en
Generate speech using Jarvis…
0:01 / 0:02I was just thinking…
Local-first by design
Your audio, transcripts and voice profiles stay on your machine. No API keys. No per-character fees. No silent uploads.
See the local APIEverything in one studio
Professional voice tools, zero compromise
Everything you need to clone voices, generate speech and produce multi-voice content — running entirely on your machine.
Near-perfect voice cloning
Multiple TTS engines for exceptional voice quality. Clone any voice from a few seconds of audio with natural intonation and emotion.
Stories editor
Create multi-voice narratives with a timeline-based editor. Arrange tracks, trim clips and mix conversations between characters.
Audio effects pipeline
Apply pitch shift, reverb, delay and compression, then save the result as a preset. Preview effects live and set defaults per voice.
Local or remote
Run GPU inference on Metal, CUDA, ROCm, Intel Arc or DirectML — or connect to a remote machine with one-click server discovery.
Audio transcription
Whisper powers accurate speech-to-text and automatically extracts reference text from voice samples for cleaner cloning.
Unlimited generation length
Generate up to 50,000 characters in one go. Text is split at sentence boundaries, generated per chunk and crossfaded seamlessly.
Any clip becomes a voice
Three ways in. One convincing clone.
Upload a clip, record from your microphone or capture audio playing on your system. Voicebox clones the voice from as little as three seconds of audio.
UploadMicrophoneSystem audio
Automatic reference-text extraction Up to 30 seconds per sample Multi-sample profiles for stubborn voices
New voice sampleReady
00:12
Click to record from your microphone.
Maximum duration: 30 seconds
Hold⌥SpaceSpeak
ListeningWhisper Turbo
Add a short welcome line before the interview starts, then keep the pace calm and leave room for the answer.
Cleaned transcript Paste into any app
Transcript ready⌘ V into any focused field
Dictation that follows you
Dictate anywhere. Paste into any app.
Hold a shortcut anywhere on your machine, speak, release. The transcript lands in a focused text field or your clipboard. Agents speak back through the same pill in any cloned voice.
Whisper, sized for every machine
Base, Small, Medium, Large and Turbo. All local, all multilingual.
Refined transcripts
An optional local model removes filler and fixes punctuation without rewriting you.
Agents speak in voices you own
Any MCP-aware client gets a voice with one tool call.
MCP voice surface
Every agent gets a voice.
One tool call — voicebox.speak — and any MCP-aware agent can talk to you in a voice you've cloned.
agent-tool-call.tsMCP
await voicebox.speak({
voice: "morgan",
text: "The render is finished.",
delivery: "calm, close-mic",
});Connectedlocalhost · no API key
Per-agent voice
Bind each MCP client to a profile. Claude Code in Morgan, Cursor in Scarlett.
Always visible
Every agent-initiated speech surfaces the voice pill. No silent background TTS.
Open protocols
MCP ships first. ACP, A2A and other tool-call primitives can share the same endpoint.
Character, not a preset
Voices with a personality.
Give any voice profile a free-form personality. Then rewrite your text in their voice, or let them compose a fresh line of their own — your cloned voice, in full character.
RewriteCompose
DN
Detective NoirJarvis · English
Persona“1940s noir detective. World-weary, cynical, every situation a metaphor for the city's underbelly. Talks like he's seen one stack trace too many.”
Compose a line
“The logs were clean, see. Too clean. Somewhere in that quiet little stack, somebody was lying.”
PreviewRoll again
Local REST API
Your local voice API.
Every engine becomes a REST endpoint on your machine. Build apps, games and voice tools with full programmatic control — no API keys, no rate limits, no per-character fees.
localhost:17493REST
curl http://127.0.0.1:17493/v1/speak \
-H "Content-Type: application/json" \
-d '{"voice":"morgan","text":"Good morning."}' \
--output morning.wavOpenAPI referenceLocal · private · fast
Games
Generate NPC dialogue on the fly, localize characters and ship expressive lines without a studio.
Apps & agents
Give your app or AI agent a voice. Real-time narration, accessibility readouts and voice replies.
Scripts & tools
Batch-generate audiobook chapters, automate intros or wire Voicebox into your own tools.
No API keysNo rate limitsNo per-character feesWorks offlineYour audio, your machine
Supported models
Pick the right model for every job.
TTS, transcription and refinement. All models run locally on your hardware. Download once, use forever.
TTS Engines
Text → speech. Voice cloning, preset voices and delivery control.
Qwen3-TTS
High-quality multilingual cloning with natural prosody and natural-language delivery control.
Chatterbox
Production-grade voice cloning with 23-language support and emotion exaggeration control.
Chatterbox Turbo
Lightweight and fast, with support for [laugh], [sigh] and other expressive tags.
LuxTTS
Ultra-fast, CPU-friendly 48kHz cloning for quick iterations.
Qwen CustomVoice
Nine premium preset speakers with natural-language style control.
TADA
Text-acoustic dual alignment for coherent long-form audio and multilingual narration.
Kokoro
Tiny 82M-parameter TTS that runs at CPU realtime with negligible VRAM.
Transcription
Speech → text. Multi-language STT for dictation and captures.
Whisper
The default multilingual ASR stack, from Tiny for speed to Large for accuracy.
Whisper Turbo
Pruned Whisper Large v3 at roughly eight times the speed for real-time dictation.
Language Models
Transcript refinement, persona replies and on-device reasoning.
Qwen3
Powers transcript cleanup, persona voice replies and the local voice I/O loop.
From the voice room
Built for people who work in voice.
Creators, engineers and narrative teams use Voicebox when the voice has to feel like it belongs to them.
“The first voice stack that felt like it belonged on my desktop instead of someone else’s server.”
“One tool call and our agent has a voice. The fact that it all runs locally is the unlock.”
“I can audition a character, rewrite the line in their voice and cut it into the story without leaving the app.”
Voicebox FAQ
What should you know before you clone?
The short answers about local processing, paid plans and giving an agent a voice of its own.
What is Voicebox?
Voicebox is a local-first voice cloning and speech studio for cloning voices, generating speech, dictating into any app and giving MCP-aware agents a voice.
Does Voicebox run locally?
Yes. The studio is designed for local GPU or CPU inference, with optional remote inference when you deliberately configure another machine. Your voice samples and generated audio do not need to leave your computer.
Is there a free Voicebox plan?
No. This Voicebox site offers paid Creator, Studio and Label plans only. Checkout opens directly on Stripe, with no payment-provider picker.
Can agents speak through Voicebox?
Yes. Any MCP-aware client can call the local voice tool and speak through a cloned profile you assign to it. Agent speech surfaces through the Voicebox pill rather than running silently in the background.
Ready when you are
Bring your voice studio home.
Available for macOS, Windows and Linux. Clone a voice, dictate a thought or give your agent a voice — without sending your audio to a server.
macOSWindowsLinuxOpen source