Breaking

Chess GM Vidit Builds a Business AI Agent Using Nemotron

July 6, 2026 at 15:04 EDT

AI Agents
Foundation Models
Open Source

Chess grandmaster Vidit Gujrathi took part in the NousResearch-run "Hermes Agent Accelerated Business Hackathon," building and submitting a deployable business agent using NVIDIA's open Nemotron model. NVIDIA highlighted the effort, thanking payments provider Stripe and NousResearch.

July 6, 2026 · NVIDIA Nemotron

A Grandmaster Built an AI Agent That Runs Its Own Business

Chess GM Vidit Gujrathi's "Kibitz" predicts the moves humans actually make in live broadcasts — and autonomously handles billing, narration, and profit-and-loss. All trained and served on a single consumer GPU. "I build. Hermes operates."

RTX 5080

One consumer GPU handles both training and inference — no cloud needed

550B

Nemotron 3 Ultra parameters (MoE) powering agent narration

$10K

Top hackathon prize + DGX Spark + Stripe credits

The gap Kibitz fills

Traditional engines always suggest the optimal move. Humans — even grandmasters — don't. Kibitz predicts the human move instead.

Optimal move

Classic chess engine
(what's best)

Human move

Kibitz, trained on 1000s of
GM games (what people do)

One agent, the whole business loop

Channel onboarding → Stripe billing → Run broadcast + narration → Track inference cost → Manage P&L → ↺ self-improving skill loop

What developers praised

A chess expert combined basic coding with AI tools to ship a practical, self-running agent — a win for local AI and for bringing such tools into the chess world.

The open question

Instability in the free Nemotron tier leaves real-world operation uncertain. Judging results — and Kibitz's rank — remain unconfirmed.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.