H Company's Holo3.1 release provides computer-control models packaged for local operation. The AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → can run on a personal computer while its model runs either on that same machine (the release gives reference numbers for Apple Silicon) or on a NVIDIA DGX Spark on the same network. This gives teams an alternative to sending model requests to a remote service.
The release also reports a concrete speed improvement on the DGX Spark: average agent-step time fell from 6.8 to 3.3 seconds compared with an FP8 baseline. H Company says that came from combining 4-bit NVFP4 quantizationShrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.Full definition → with agent-agent harnessThe scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.Full definition → optimizations developed with NVIDIA, so it should be understood as a result for the complete tested setup.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
quantization — Shrinking a model by storing its numbers less precisely — like rounding — so it runs faster and fits on smaller hardware, at a small quality cost.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
