A few days ago we moved Chervil's wake word to openWakeWord — free, open-source, fully on-device, no API key. The one catch: saying the literal "Hey Sprig" meant training a model yourself and loading it as a custom file. Convenient if you'd done the Colab run; a speed bump if you hadn't.
That speed bump is gone. Chervil v0.1.4 bundles a trained "Hey Sprig" model and makes it the default. Install, turn on listening, and the word already works — no notebook, no GPU, no .onnx to load.
How to use it
Open Settings → Listening and flip listening mode on. The wake word is already set to Hey Sprig, so that's it — say "Hey Sprig" and the quick-ask bar pops up, ready for a spoken request. Grant the mic permission the first time and you're hands-free from there.
One tip from our own testing: say it in a normal, conversational tone. If you've been trained by smart speakers to announce the wake word — the clipped, emphatic "HEY. SPRIG." — ease off. openWakeWord learns from naturally spoken phrases, so a relaxed, everyday delivery is what it recognizes best.
Nothing about the model changed underneath
This is the same engine as before, just pre-loaded: detection runs entirely on your machine, your mic audio never leaves the device until you actually speak a command, and there's still no account and no key. If you'd rather use a different word, Hey Jarvis, Alexa, and Hey Mycroft are still one dropdown away — and you can still train and load your own custom word if you want something else entirely.
Get it
It's in the latest Windows build. Download Chervil v0.1.4, turn on listening, and just talk to the web.