Hevolve Hive Agentic Runtime The AI-native operating system for every device, from your computer to embodied AI. Local-first, federated, OpenAI-compatible. The most capable AI now lives inside a handful of organisations that can put enough compute in one building. That concentration is an architectural choice rather than a law of physics, and it decides what the models will refuse, what they cost, and who gets to read what you type. Two different locks are usually treated as one. There is the capital lock, the compute it takes to pretrain a frontier model, and there is the silicon lock, the fact that most of the stack assumes one vendor's cards and one vendor's toolkit. The second is the one we refuse outright. Local inference here runs through llama.cpp on GGUF weights, so CUDA, ROCm, Metal, Vulkan and plain CPU are all first-class. vram_manager treats "CPU-only, Metal, broken nvidia-smi" as ordinary cases rather than failures, and probes rocm-smi next to nvidia-smi. An Apple laptop, an AMD card and a machine with no GPU at all are supported paths, not degraded ones. Nothing in the delivery path requires a particular company's silicon. On the capital lock, we are precise rather than loud: you cannot pretrain a frontier model without a datacenter, and we do not claim you can. What we claim is that the intelligence people actually use does not have to be delivered from one, and that the gap between open weights and closed ones has been closing faster than the compute gap has been widening. HART OS runs models on the device, and when a node cannot handle something it recruits a peer directly, with no broker and no company in the middle. We do not shard a model across the network either. Layer-level parallelism over consumer links is a bad idea and we do not attempt it. A peer serves a whole request instead, so a machine with a bigger model answers what a smaller one could not, which tolerates latency in a way tensor parallelism does not. The things we have not solved ...
First seen: 2026-07-26 20:05
Last seen: 2026-07-26 20:05