meldivo
v2.4.7
Published
Always-on local voice hub for coding agents: talk hands-free to your Pi, OpenCode, and Claude Code sessions from any browser or phone. Private on-device speech (Parakeet + Kokoro). No cloud speech APIs.
Maintainers
Readme
Get started in three steps
1. Install
npm i -g meldivo2. Start
meldivo startmeldivo runs in the background from now on and prints your private hub link with a QR code.
3. Talk
Open the link in any browser, or scan the QR code. Every Pi, Claude Code, and OpenCode session on your machine is already there. Pick one, or start a new chat, and just speak.
That's it. No API keys, no plugins, no cloud.
Why meldivo
- Works with the agents you already use. Pi, Claude Code, and OpenCode sessions show up automatically, each running with its own model and credentials.
- Private by design. Speech recognition and synthesis run locally on your
CPU. The hub listens on
127.0.0.1only and every link is protected by a secret token. - Never in your way. If a session is open in a terminal, voice continues it as a fork, so nothing collides with what you're typing.
- Anywhere you are.
meldivo remoteputs the hub on your phone over HTTPS, and your other computers can join one hub with a one-time code. - Bring your own voice. Point it at your own speech server (Breeze TTS 2, vLLM-Omni, Kokoro, Whisper, and more) for bigger or more realistic models.
Requirements
Node.js 22+, macOS or Linux, and at least one of Pi, Claude Code, or OpenCode. Speech models (about 650 MB) download once on first use.
Documentation
Commands, phone access, multi-machine hubs, speech servers, configuration, and security are all covered in the documentation.
Acknowledgements
Some of the methods used in meldivo were inspired by, and adapted from, the ones described in How I Built an End-to-End Local Voice Agent, and Made It Fast by Codacus and used in his Pithagoras project.
License
MIT. See LICENSE and third-party notices.
