OMG So cool!
Jensen closed his GTC Berlin 2026 keynote with this Hermes Agent demo as the finale.
He introduced it by saying the way to build an agent is to ask an agent to build it for you, so he installed and set up a customized Hermes in one prompt.
Then he used Hermes, running on @nvidia RTX Spark, to create a simulated training environment for the @huggingface Microduck. Hermes planned an obstacle course with Nemotron and had GPT-6 Astra build it using FreeCAD, Omniverse libraries, Blender, Newton, and Isaac Lab. Hermes trained thousands of simulated ducks on that course until the validated policy was ready to run on the real robot in the physical world.
At Nous, we believe open source is a practical tool as much as a philosophical ideal. It gives users the ability to build powerful, complex workflows chaining numerous local+cloud models, devices, and tools themselves instead of waiting for a company to enable them.
Ohhh! GUI now. No more Browser App. Much better Tailscale integration working, PC use on Sunrise now to help with Lab setup. Kinda fun. You download and install Sunrise sign in to huggingface in Sunrise. Download the model that fits yor PC. Mount the model. Sign in to Tailscale in Sunrise. Then you enable computer use and Sunnie will help you with the rest setting up the Lab. Connecting you to your Openclaw or Hermes Agent on other PC's.
I’m developing SunRise, a local AI inference engine built on Strata, using an AMD FX-8350 system with two RTX 3060 12 GB cards.The CPU dates to 2012. Working with its limitations is a big part of the project: an AVX-only build, DDR3 memory, and two separate GPU memory pools.I’m documenting what works, what fails, and what I can measure on this setup. If you’re interested in local AI on older hardware, follow along on my Substack. substack.com/home/post/p-221…
Pulled this repo Used Open AIs Daybreak to find all the security issues and there were quite a few Were going to fix them. Amazing work has been done here. I think we can make it great. Going to customize it for my very old harware. DDR3 is a big hint. Its already running on the dino at 30-40 TOK/s 256k context. github.com/Niko1221/Strata
1,216 training steps. 660 graded answers. Citation compliance improved sharply—but project-behavior pass counts did not. The candidate failed our gate, so Hermes stayed unchanged. Full results: failures, speed, long-context checks and lessons learned. #LocalAI








