VocaGateway v0.1.0 release

optional self-hosted speech-to-text

vocagateway

your hardware, shared speech-to-text.

try v0.1.0 view the source

beta · macOS · Linux · Docker Compose · AGPL-3.0 · part of vocahq.com

  • not on-device
  • pair with a QR code
  • no Voca account
record✦send✦transcribe✦your host✦

part of vocahq.com

VocaGateway is optional infrastructure for the Voca family. It is not a client app and it is not on-device. On-device speech-to-text stays on the phone or computer; this path is the one you stand up when you want shared or larger compute. See the full map on vocahq.com.

the WebUI

VocaGateway WebUI Overview. Ready for dictation.
Overview Real WebUI capture. Ready for dictation.
VocaGateway WebUI Pair and test. Example QR from a stopped local host, not a live token.
Pair and test Real WebUI capture. Example QR from a stopped local host, not a live token. After you run a host, scan the QR in your own WebUI.
how it works

run it once.
pair the client.

Set the gateway up on hardware you control. Clients send a bounded recording. A speech-to-text model on that host returns the transcript.

1once

run it on a host you own

Native on a Mac or a Linux box, or Docker Compose on Linux amd64/arm64. There is no Voca-hosted service and no installer to download.

2pair

scan the pairing QR

After you open the WebUI with the bearer token, Pair & test shows a QR. VocaPhone on iPhone or Android reads the gateway URL and token from it.

3speak

get text back

The client records. Audio travels to this host. The selected model runs here and the transcript goes back to the app. Desktop clients are planned; VocaPhone is the current consumer.

where speech-to-text runs
privacy is a route

on the host you run.
not a Voca cloud.

Gateway mode is not on-device. When a client is pointed at VocaGateway, audio leaves that device and travels to the machine you configured. The gateway runs a local speech-to-text model and returns text.

01
Trusted LAN

Fine for a home or lab network you already trust. HTTP does not encrypt the token or the recording.

02
Private encrypted network or HTTPS

Use Tailscale Serve, another encrypted network, or a reverse proxy with a real certificate when the path leaves that LAN.

03
Not the public internet on :8765

Do not publish port 8765 to the open internet. That listener carries the bearer token and the audio.

get started

run it from source
or from Compose.

Status is Beta. There is no packaged installer. Run Compose or native from the v0.1.0 tag, not from main. After the process starts, open http://127.0.0.1:8765/, enter the token, download a model, and wait until Overview says Ready for dictation.

native macOS

Apple silicon host

brew install ffmpeg whisperkit-cli whisper-cpp
uv sync --all-groups --extra engines --extra apple
uv run vocagateway

native Linux

desktop or home server

sudo apt install ffmpeg
uv sync --all-groups --extra engines
uv run vocagateway

Docker Compose

Linux amd64 / arm64

cp .env.example .env
printf 'VOCAGATEWAY_TOKEN=%s\n' "$(openssl rand -hex 32)" >> .env
docker compose up --detach --build

command name: vocagateway is the CLI. Deprecated vocaphone-server and related aliases still work for one cycle. The product is VocaGateway. Full host notes, Tailscale, and reverse-proxy setup live in the README and deployment guide.

the Voca ecosystem

same privacy bar.
different machines.

Start at vocahq.com for the directory. Each product keeps its own site, status, and source.

family home

vocahq.com Private speech-to-text across the machines you own.
iPhone + Android beta / testflight

VocaPhone

The current gateway client. Android has a public beta. iPhone is a public TestFlight beta capped at 1,000 seats, or an iOS 17+ source build. On-device is the default path; this gateway is optional.

vocaphone.vocahq.com TestFlight source
Linux available now

VocaLinux

System-wide voice typing for real Linux desktops. On-device speech-to-text on X11 and Wayland. Gateway shipping from the desktop app is planned.

vocalinux.com source
macOS beta

VocaMac

A native menu bar app for Apple Silicon. Hold a hotkey, speak, and text appears at the cursor through an on-device speech-to-text model.

vocamac.com source
Windows beta

VocaWin

Unsigned v0.1.0-beta.1, not a store listing. Windows SmartScreen will warn. It does not expose a gateway mode today; desktop embed is planned.

vocawin.com source
questions, answered

the useful bits.

Is VocaGateway on-device?

No. On-device means the speech-to-text model and the audio stay on the phone or computer after the model download. Gateway mode is different: configured audio travels to the host you run. Do not collapse those two paths.

Do I need this for VocaPhone?

No. VocaPhone can transcribe on the phone after you download a model. Use VocaGateway when you want shared hardware, a larger model, or one service for more than one client.

What can pair with it today?

VocaPhone on iOS and Android. Linux, macOS, and Windows desktop apps are planned to start this same headless server later. There is no Voca account.

Where should I expose the port?

A trusted LAN, Tailscale Serve (or another encrypted private network), or HTTPS on a reverse proxy. Never publish port 8765 to the public internet. HTTP does not encrypt the bearer token or the recording.

Is it free and open source?

Yes. The repository is github.com/VocaHQ/vocagateway, licensed under AGPL-3.0. The rest of the family lives under the VocaHQ organization.

beta · self-hosted · optional

run it on hardware
you already have.

Start from the v0.1.0 tag, run the process, pair a phone. If you only need on-device dictation, skip this and pick a client on vocahq.com.

view the source open vocahq.com