-
v1.5.0 Stable
released this
2026-09-04 20:29:32 +00:00 | 48 commits to main since this release- 100% Self-Hosted Nahalo AI Suite & Zero-External-Dependency Architecture:
- Fully decoupled visual card recognition and embedding pipelines from external third-party hosts (HuggingFace).
- Added support for FORGEJO_ASSETS_BASE_URL environment variable and settings.forgejoAssetsBaseUrl to synchronize all AI model weights (cornelius.onnx, milo.onnx, and nahalo-catalog.npz) directly from self-hosted Forgejo releases.
- Added 1-click "Sync All from Forgejo" action in Settings > Scanner & AI Models.
- Standalone Milo Neural Vector Indexer CLI:
- Implemented standalone CLI script index-vectors.ts
(npm run index:vectors) to encode Scryfall cards into 128-dimensional Milo neural vector embeddings with multi-threaded CPU acceleration (8 threads), concurrent image downloads, and atomic batch checkpointing. - Added --bundle-all (-b) flag to consolidate upstream base cards and supplemental cards into a single, complete, self-contained nahalo-catalog.npz (121,369 total Scryfall paper cards, 75.9 MB).
- Unified Standalone Vector Catalog Loader & Smart Auto-Pruning:
- Enhanced CardScanService to load nahalo-catalog.npz as a standalone complete vector index in a single contiguous Float32Array memory buffer.
- Web Worker & ONNX Runtime Web Diagnostics:
- Added live browser Web Worker benchmark in Settings to test client-side WASM SIMD acceleration, frame processing latency, and corner detection responsiveness.
Downloads
-
Source code (ZIP)
5 downloads
-
Source code (TAR.GZ)
1 download