AMD Strix Halo 395 + Hermes + Hindsight = all local, works great!

[BBooDad]

Just wanted to complement the Hindsight team. Hermes with Qwen 3.6 does a great job... until it doesn't and then have to restart it. Half the time it hasn't written to the markdown files it is supposed to. And half the time it doesn't remember what it is supposed to do on initialization. Enter Hindsight. On my AMD Strix Halo, I just happen to have fat happy NPU, running Qwen3-VL, so it can take scanned documents anytime and process them on to my NAS, and run Home Assistant jobs. None of which is is very often, relatively speaking. So I gave it Hindsight to manage, too. All that running on the NPU: 2-4 watts of power, when it is crunching numbers, or whatever magic it does.

Now, Hermes initializes correctly every time, and remembers what it was doing or did and how. 500% more useful.

Thanks vectorize.io!

[sergiopadure]

Which LLMs are you using for Retain, Recall, Reflect?