Why it exists

Voice AI got good, then it got rented: the best cloning quality is locked behind per-character cloud pricing, accounts, and sending your voice — among the most identifying data you have — to someone else's servers. Open TTS engines exist, but stitching them into a usable desktop workflow is a Python-y weekend project, not a product. Voicebox is that product: a polished, MIT-licensed desktop studio that runs multiple state-of-the-art engines locally, keeps your voice samples on your disk, and layers dictation and agent integration on top.

Main features

• Voice cloning from samples — import a voice, add multiple samples, get a reusable cloned voice; shipped presets range from "composed British AI assistant" to "commanding intensity".
• Seven TTS engines — switch engines per job without leaving the app; smart caching avoids regenerating identical output.
• Local or remote inference — everything can run entirely on your machine; point it at remote endpoints if your GPU is humble.
• System-wide dictation — dictate into any app, positioning Voicebox against cloud dictation tools like WisprFlow.
• MCP integration — talk to agents in voices you own: Voicebox exposes itself over MCP so assistants can speak through it.
• Capture & studio workflow — recording/capture built in, with a New/MCP/Models workflow for managing engines and voices.
• Cross-platform — macOS, Windows and Linux builds from one codebase; v0.2.0 is current.

Who it's for

• Creators who want ElevenLabs-class cloning without per-character billing or uploading their voice to the cloud.
• Privacy-conscious users: voice is biometric data, and with Voicebox it never has to leave your disk.
• Developers and tinkerers wiring speech into local agent setups via MCP.
• Anyone who dictates — emails, notes, code comments — and wants it fast, free, and offline-capable.

How it compares

• ElevenLabs — the quality benchmark, cloud-only with usage pricing; Voicebox trades a bit of polish for full local control and zero marginal cost.
• WisprFlow / cloud dictation — slick and multi-device, but audio leaves your machine; Voicebox dictates locally.
• Piper / Coqui / raw engine CLIs — engines, not products; Voicebox is the multi-engine studio UI (import, cache, presets, MCP) wrapped around them.
• Murf / Descript studio voice — content-production suites with subscription seats; Voicebox is a free desktop tool focused on cloning + dictation, not video editing.

Notes

• From Jamie Pine (Spacedrive, Discover.me) — same independent-developer pedigree, same public repo, 57k stars and growing fast.
• Quality varies by engine and sample count; cloning a voice from short or noisy samples yields exactly the artifacts you'd expect. Use voices you have rights to.
• Local inference wants a decent GPU; the remote-inference option exists precisely for laptops that can't keep up.
• MIT license — self-build, fork, or embed it; donations accepted on the site.
• Desktop-only today; there's no mobile app, and multi-device sync of your voice library is DIY.

Useful links

• Official site: https://voicebox.sh/
• GitHub: https://github.com/jamiepine/voicebox
• Download: https://voicebox.sh/ (macOS, Windows, Linux)
• Docs: https://voicebox.sh/docs