https://news.ycombinator.com/item?id=47155258
Finally got around to dusting off a old setup and I am thrilled with the results. I'm using a 200$ used GPU and 11year old CPU. Basically a retired game rig. I can now host a private model to my entire family that is just as capable as the things they were using it for without worrying of all the private information leaking out. While I am extremely impressed on the outputs of the local model, I am not to concerned about its quality. It's fine. What's more exciting is it's a path forward to data sovereignty and ownership.
Great job on the privacy policy.
https://locallyai.app/privacy-policy.html
Setup was easy.
As hardware improves, and local models improve...this is the way.
The connection is powered by tailscale, long-term I would like to see other implementations of this wire to diversify how these devices are connecting.
Iroh, recently posted here comes to mind... But in general, a stream of text is not a heavy payload and shouldnt need to rely on a loss-leader from any one place.