Local & on-device models
Private, free AI that runs entirely in your browser on your own GPU — no account, no data leaving your device.
Parley can run small language models directly on your device using WebGPU — the same GPU that powers your games and video. Nothing is sent to a server: the model downloads once into your browser, then every reply is generated locally.
Why use a local model?
- Private — your messages never leave your device. Nothing is stored or sent anywhere.
- Free — no credits are spent, because there’s no cloud call.
- Instant to try — the demo at parley.live runs without an account.
What local models can’t do
On-device models are small — think TinyLlama and Llama 3.2 1B. That keeps them fast and private, but they trade away capability:
- Text only. They can’t read images, generate images, or make video. Attach an image and Parley switches to a cloud vision model for that turn.
- Shorter memory and simpler reasoning than the cloud models.
- First use downloads the model — a one-time wait while it caches in your browser.