v0.1.0 - Initial Release

HAT: Hierarchical Attention Tree

Initial public release of HAT - a novel index structure for AI memory systems.

Highlights

• 100% recall on hierarchically-structured data (vs ~70% for HNSW)
• 70x faster build times than HNSW
• 3.1ms query latency at 60K+ tokens

Installation

Python:

pip install arms-hat

Rust:
[dependencies]
arms-hat = "0.1.0"

Links
PyPI
crates.io
Documentation
Paper (PDF)

What's New

• Core HAT index with beam search queries
• Session/document/chunk hierarchy management
• Persistence (save/load)
• Consolidation phases for index maintenance
• Python bindings via PyO3