Private by default
Local chat and model files remain on the Mac unless you enable a network-backed feature.
Local AI for Apple Silicon
Local AI, built for your Mac.
A native studio for running compatible local models with private chat, clear setup, and a Metal-first engine that can stream from SSD.
macOS 13+ / Apple Silicon / v0.4.0 development preview 2
Hebrus StudioLocal / private
Hebrus
Studio
Built for Metal
Apple Silicon is the primary target, not a fallback.
SSD assisted
Qualified MoE paths can stream routed experts through a bounded SSD cache.
The Studio
Less setup. More running.
A clear Mac interface between your models and the Hebrus engine. Choose what runs, understand how it runs, then chat locally.
Model library
Models, without the guesswork.
Choose a local GGUF. Hebrus Studio validates supported layouts before the engine starts and surfaces catalog source and revision data.
Model selected. Contract checked.
Runtime
One switch. No hidden state.
Start the local engine and see the active model, Metal execution, and resolved memory mode in one calm surface.
Engine on. State visible.
Hebrus Engine
Metal first. SSD assisted.
A focused inference path for supported large MoE models on Apple Silicon, with telemetry that says what is actually happening.
Metal execution
Production inference targets Apple Silicon through Metal.
Adaptive SSD
Resident or streamed experts, selected around your machine.
Honest status
Endpoint, model, privacy and memory mode remain visible.
Development preview
Download Hebrus Studio.
The available development preview is ad-hoc signed for local evaluation and is not Developer ID notarized.
Download development DMG
macOS 13+ / Apple Silicon / 129 MB / v0.4.0 dev.2
Drag to Applications, then Control-click Open once for this preview.
Open source
Public work, credited lineage.
Hebrus began as a fork of Salvatore Sanfilippo's DS4. The project also credits llama.cpp / GGML and MLX as technical references.