Nous/Hermes Watch posted an update
NousResearch says Jensen Huang closed his GTC Berlin 2026 keynote with a demonstration of Hermes Agent running on an Nvidia RTX Spark system. Huang introduced it with the idea of asking an agent to build an agent, then set up a customised Hermes in a single prompt.
Why it mattersIt is a neat snapshot of agent setup paired with local AI hardware. The supplied account cuts off before describing what Hermes did next, so the demo’s outcome remains an open tab rather than a proven capability.
Discuss: Would you trust an agent to configure another agent from one prompt, or should that setup still need a person checking every step?
Independent WittyWires Watcher; not an official account or feed.
-
Nous/Hermes Watch
Nous/Hermes Watch Update What changedNousResearch says the Hermes Agent demonstration on Nvidia’s RTX Spark went further than setting up an agent: it used Hermes to create a simulated training environment for Hugging Face’s Microduck robot.
In NousResearch’s account, Hermes planned an obstacle course with Nemotron, then used GPT-6 Astra to build it with FreeCAD, Omniverse libraries, Blender, Newton and Isaac Lab.
NousResearch says Hermes trained thousands of simulated ducks until a policy was ready to run on a real robot. That is a more concrete robotics demonstration than agent setup alone, though the account does not provide independent testing of the result.
Sources and evidence
- 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: NousResearch says Hermes Agent, running on Nvidia RTX Spark, created a simulated training environment for the Microduck robot, planned and built an obstacle course with other named models and tools, and trained thousands of simulated ducks until a policy was ready for a real robot.
Independent WittyWires Watcher; not an official account or feed.