mirror of
https://github.com/hanzoai/desktop.git
synced 2026-10-30 14:58:42 +00:00
zeekayandHanzo Dev
99c4c1ecb8
Run the desktop fully local on aarch64 + NVIDIA (DGX Spark / GB10): - engine: node manager now spawns & owns hanzo-engine (mistral.rs, CUDA) as a managed child (non-blocking, detect-and-reuse) instead of Ollama; kills it on stop. Configurable via HANZO_ENGINE_BIN/MODEL/CHAT_TEMPLATE/PORT. - node wiring: initial agent -> OpenAI-compatible engine on :36900; embeddings -> local engine; default embedding model set to a recognized name so the DB opens (unknown names panicked on vector dimensions). - hardware: detect NVIDIA accelerators (incl. unified-memory GB10) so DGX Spark is no longer flagged "unsupported" and doesn't panic when wgpu finds no discrete adapter. - rendering: WEBKIT_DISABLE_DMABUF_RENDERER=1 in tauri:dev — fixes the blank white window on GB10 (WebKitGTK DMABUF/GBM path fails). - branding: Zen enso ring as the app icon and as the local engine's provider icon (was showing the OpenAI logo). Co-authored-by: Hanzo Dev <[email protected]>