Sirene

Appearance

Getting Started ​

Desktop App ​

Download the latest release for your platform from the releases page:

PlatformFile
macOS (Apple Silicon)sirene-desktop-macos-arm64.zip
Linux (x64)sirene-desktop-linux-amd64.tar.gz
Windows (x64)sirene-desktop-windows-amd64.zip

The download is a few tens of megabytes. On first launch, Sirene downloads a standalone Python and the inference service into ~/.sirene (a few hundred megabytes); the Local inference server shows as offline in Settings → Inference servers until that finishes. TTS backends themselves are installed on demand when you install a model. Logs are in ~/.sirene/logs/desktop.log.

Docker ​

The quickest way to run Sirene as a server:

bash
curl -sSL https://raw.githubusercontent.com/KevinBonnoron/sirene/main/install.sh | bash

Supply-chain note. The bootstrap script is fetched from main. For production deployments review the latest release and replace main with the corresponding tag once a release that supports your installation mode is published.

See the Docker guide for more options.

First Launch ​

  1. Open the app (http://localhost for Docker)
  2. Create the administrator account in the setup wizard
  3. Go to the Models page and install the model of your choice (e.g. Kokoro v1.0)
  4. You're ready to generate speech

Adding more inference servers ​

If you have a separate Linux machine with a GPU, or a hosting provider such as Railway, you can add it as an inference server:

  1. In Sirene → Administration → Inference servers → Add server. The dialog shows an install command carrying a one-hour registration token.
  2. Run that command on the machine, or deploy the Docker image with the listed environment variables.
  3. The worker registers itself with Sirene when it starts. It appears in the list within a few seconds, health-checked.

Nothing needs to be pasted back: the worker sends its own URL and auth token during registration. The manual form (URL + token) is still available under Add manually for workers that are already running.

Keep the auth token secret. It grants full control of the inference server. Do not paste it into chats, screenshots, or logs, and rotate it if you suspect it has leaked.

Rotating a token. Delete auth_token on the inference host and rerun the installer with a fresh registration token: the worker re-registers with the new value. Without registration, update the matching entry in Sirene by hand; until then every request to that inference server returns 401.

See the Docker guide for details.