Everything needed to run ships as a single artifact.
Private runtime. Public proof.
A different way to run AI on real hardware.
HOBS rethinks how models are stored, loaded, and executed — with a focus on compactness, transparency, and practical deployment without specialized infrastructure.
Current framing
Smaller artifacts, self-contained loading, commodity hardware, and visibility into what the model is actually doing.
Designed to run usefully without requiring specialized accelerators.
Compression that preserves what matters instead of discarding it uniformly.
You can see what the model is doing while it works.
Evidence
Why this is worth paying attention to.
Format and execution are co-designed
The way HOBS stores a model and the way it runs are built together — size, structure, and runtime behavior stay aligned by design.
Compression preserves what matters
Rather than uniformly shrinking everything, HOBS keeps the important structure intact while staying within predictable size budgets.
CPU inference
Fast enough to matter
Large model handling
Single-file load on desktop-class hardware
Training visuals
Structure emerges visibly
Representation
A different way to look at model internals
Demo
Built to show the process, not just the output.
Live training visualization
Watch the model's internal structure take shape as it learns.
Runtime dashboard
A live overlay showing what the model is doing as it runs.
Self-contained demos
Load and run a model from a single file, no external dependencies.
Deployment
Simple hosting path for a focused public drop.
Site
Static site for benchmarks, visuals, writeup, and access points.
Demo
Hosted runtime for controlled, interactive model demos.
Evidence bundle
Logs, metrics, video, and artifact descriptions collected in a release-oriented structure instead of buried across repos.