VoxSign is a local-first voice AI platform. Hold to talk, let your machine listen, and only flip on the cloud when a session actually needs it. Your voice stays where it belongs — with you.
A calm voice layer between you and your devices — built on the open Harness runtime, not a black box.
Press the big waveform button and speak. Release to send. No menus, no typing — just voice, the way you already talk.
Run parallel conversations, each with its own role and domain. Work, personal, experiments — never bleed into one another.
One tap moves a session between your machine and the cloud. Green means on-device, red means routed out — you always know.
Threads title themselves from your first turns. No more "New chat 47" — just open the one you mean.
The brains ship as Apache-2.0 software: session state, model routing, local/cloud transport, and the hold-to-talk protocol. Point it at your own model, your own endpoint, or your own server.
Apps, runtime, and server live in the Voxsign GitHub organization — every line under the permissive Apache-2.0 license. Fork it, audit it, run it your way. Voice infrastructure should not be a moat.
Android builds are published straight from GitHub Actions. iOS is rolling out via TestFlight.
Download the latest APK from the VoxSign-Android releases page — built and signed by GitHub Actions.
Source: github.com/Voxsign/VoxSign-AndroidTestFlight openings will be announced here. The iOS source is already public — build and run it yourself from the repo.
Repo: github.com/Voxsign/VoxSign-IOS