All notable changes to sentinel-reverse will be documented in this file.
- Initial open-source release extracted from Innora-Sentinel
- MLX Engine: Local LLM inference with 6 analysis task types (decompile, function naming, vulnerability detection, algorithm detection, obfuscation restore, type inference)
- MPS Accelerator: PyTorch Metal GPU acceleration for binary code embeddings, cosine similarity search, and function name prediction
- Iterative Analyzer: Multi-round confidence-driven analysis with 4 strategies (full_reverse, quick_decompile, security_audit, algorithm_recovery)
- Model Voter: Multi-model voting with majority, weighted confidence, and best-of strategies
- CLI: Complete command-line interface with 4 analysis modes (quick, standard, exhaustive, security)
- Checkpoint/Resume: Per-phase incremental saving with
--resumeflag - Binary Format Detection: 9 format magic byte signatures (ELF, PE, Mach-O, APK, DEX, etc.)
- .NET Bundle Extraction: Single-File Bundle parsing with embedded DLL extraction
- Report Generation: Markdown + JSON output with full analysis details
- GPU Memory Management: Staged MPS/MLX transitions to prevent unified memory conflicts
- Model Auto-Discovery: Searches standard paths for locally available MLX models
- Multi-language README: EN, ZH, JA, KO, VI, ES, FR, DE, PT
- Test Suite: 50+ tests covering CLI, engine, voter, accelerator components