Brody Brooks

20yo founder building ThermoAgent, patent pending AI infra that routes queries to the optimal model for the task, delivering reliable intelligence per dollar.

[Brody Brooks @thermoagentai]

ThermoAgent Inc. is excited to announce its membership in the NVIDIA Inception program.
We’re building AI infrastructure around a simple goal: deliver more reliable intelligence per dollar, while simultaneously improving the efficiency, control, and reliability of AI workloads.
Joining the NVIDIA Inception program is an exciting milestone as we continue building, benchmarking, and pushing ThermoAgent toward real-world deployment.
A lot more work ahead. Back to building.
#NVIDIAInception
© 2025 NVIDIA, the NVIDIA logo, and NVIDIA Inception are trademarks and/or registered trademarks of NVIDIA Corporation in the U.S. and other countries.

[Brody Brooks @thermoagentai]

@NVIDIAAIInfra @AVEVAGroup @EmeraldAi_ @ETAPsoftware @GEVernova @hitachienergy @SchneiderElec @Siemens_Energy When the GPU gets faster but someone still routes every prompt to the biggest model available.

[Brody Brooks @thermoagentai]

@ORNL We reached out to ORNL from Kingsport, TN. We want to test whether ThermoAgent’s AI control layer can extend into a datacenter control plane using utilization, queue state, power, energy and thermal telemetry to make better workload placement decisions w/o sacrificing reliability

[Brody Brooks @thermoagentai]

@NVIDIAAI @JGrandillon If agents are becoming the software layer of every company, they’ll need an execution control layer underneath them, deciding which model/compute path to use, verifying results, recovering from failures, and tracking the economics. That’s where we fit.

[Brody Brooks @thermoagentai]

@nvidia Especially when the math becomes a control system for compute. That’s the fun part.

[Brody Brooks @thermoagentai]

@CrusoeAI @ollama I’ve built a provider-agnostic execution-control layer that routes workloads across cloud, local, and model environments based on quality, cost, latency, verification, and reliability. Crusoe looks like a natural infrastructure target. Open to testing it as one?

[Brody Brooks @thermoagentai]

@NVIDIAAI @NVIDIAGTC Looking forward to Jensen’s keynote. Curious to see what gets announced for AI infrastructure and accelerated compute.

[Brody Brooks @thermoagentai]

@SpaceX Birthing Satellites

[Brody Brooks @thermoagentai]

@thejustinwelsh Nature of the beast.

[Brody Brooks @thermoagentai]

@pmitu Sales is a different kind of animal.

[Brody Brooks @thermoagentai]

@Windows @NVIDIARTXSpark The one with the most local AI headroom. I’ve got routing experiments to run. 👀

[Brody Brooks @thermoagentai]

@NVIDIARTXSpark @ASUS @Dell @HP @Lenovo @Microsoft @msiUSA Every fire starts with a Spark, just like every brand starts with an idea. 🔥 #ShareYourSpark

[Brody Brooks @thermoagentai]

@NVIDIARTXSpark @ASUS I’d use it as a portable local inference node to test routing, verification, recovery, and model selection across cloud + local workloads. #ShareYourSpark

[Brody Brooks @thermoagentai]

@bryan_johnson lol

[Brody Brooks @thermoagentai]

@nvidia You should look at the commercial Volvo Truck market too.

[Brody Brooks @thermoagentai]

@NVIDIAAI @AIatMeta 816B is impressive, but not every coding task needs 816B parameters and a 1M+ token window. The interesting problem is deciding when that level of capability is actually justified, and when a smaller model can deliver the same result for less.

[Brody Brooks @thermoagentai]

@beffjezos Inference is only one piece of the puzzle.

[Brody Brooks @thermoagentai]

@magdovitz @wolfejosh @goformic @physical_int @samanfarid @LorimerVentures @SJCizmar @Farshchi They seem to like their job.

[Brody Brooks @thermoagentai]

@_mctrinh That’s why I built ThermoAgent. If you use multiple LLMs, it pick the cheapest model that still meets the required quality/performance, and avoid wasting tokens on capability the task doesn’t need. Reliable intelligence per dollar.

[Brody Brooks @thermoagentai]

@nvidianewsroom More compute helps, but safety also needs better execution evidence. If only there were a way. 👀