The Framework Desktop still is the best AI Token per Watt computer out there.
Local inference is getting fantastic and there’s no rate limits and no subscription.
With my heavy usage this machine pays for itself every couple months.
The water is warm, jump on in:)
@LogicSnse I use Hermes Agent, I was an early adopter of both the claw and Hermes and ended up sticking Hermes. (We even have a nous research tile on our site)
I was using qwen3.6-27b-MTP but recently switched to 35b-a3b-MTP it’s smart enough for most of my agentic flows and is very fast
@thegratefulphil@jawwwn_@elonmusk@PeterDiamandis Of course! It’s a team effort for sure. But no one involved is dumb. And this honestly seems in some ways harder of a thing to pull off than a clean fab.
The second Hermes Agent by @NousResearch came out our offices @FrameworkPuter desktops have been doing nothing but hosting LLMs to run it.
Owning 100% of your own agent infra is the coolest feeling, and if you are running one of the latest qwen models then you are not compromising on intelligence at all.
At this point this Desktop is a no brainer for anyone working in AI, I prefer my own instance of Qwen in the HermesAgent harness over any other LLM in any other harness.