Local Server Setup

RIFT supports local servers for private offline transcription and even LLM transformations.
rift-local is the recommended server: it handles model downloads, server configuration, and supports multiple backends.

Quick Start

uv tool install rift-local && rift-local serve --open

That's it! Your browser will open with the Local source pre-selected. Click Enable Voice Input to start.

Don't have uv?

brew install uv

Or see the uv installation docs for other platforms. Requires Python 3.10+.

Alternative: install with pip

python3 -m venv .venv && source .venv/bin/activate
pip install rift-local

Model	Params	Disk	Notes
`nemotron-en`	0.6B	600MB	Best accuracy (int8)
`zipformer-en-kroko`	~30M	68MB	Lightweight, fast

Model	Params	Disk	Notes
`moonshine-en-medium`	245M	190MB	Default; best moonshine accuracy
`moonshine-en-small`	123M	95MB	Balanced
`moonshine-en-tiny`	34M	26MB	Fastest