DictatorFlow Desktop and Local Inference Are Now Free
DictatorFlow's native desktop app is now a direct, free download for Windows, macOS, and Linux. There is no account wall, subscription check, or per-minute charge when you run a model on your own machine.
- > The native DictatorFlow desktop binary.
- > Unlimited local transcription with a compatible local model.
- > CPU inference and supported NVIDIA CUDA or AMD ROCm builds.
- > Offline use where recordings stay on your computer.
One public install path
The download page is public and the standard installers no longer prompt for cloud credentials. Existing cloud configuration is preserved during upgrades, but a fresh installation starts in local mode and lets you choose a model.
# macOS or Linux
curl -fsSL https://dictatorflowstatic.dictatorflow.com/install.sh | sh
# Windows PowerShell
irm https://dictatorflowstatic.dictatorflow.com/install.ps1 | iexLocal model choices
Parakeet ONNX remains the recommended local transcription path. DictatorFlow can use NVIDIA CUDA, an ONNX Runtime build with AMD ROCm, or CPU fallback. GPU acceleration is optional: local inference does not become paid just because it runs on different hardware.
Microsoft's VibeVoice-ASR-BitNet is also available as an experimental CPU path on Linux. It is free and private, but it is not our default because the first DictatorFlow corpus measured 60.45% average WER. The engineering report explains the benchmark and the opt-in installer.
DictatorFlow's hosted transcription models, managed provider fallback, subscriptions, and developer API usage remain paid cloud services. They are optional conveniences, not licenses for the desktop app.
Opening the source
DictatorFlow source now carries an Apache 2.0 license. Contributor and security guidance, third-party notices, asset provenance, secret checks, and safer assistant boundaries are part of the release work. Public release artifacts still pass a final asset, credential-history, deployment-default, and trademark review.