v0.10.0 — Retrieval Feedback Loop

What's New in v0.10.0

Retrieval Feedback Loop (Phase 1 of v1.0 Roadmap)

• FeedbackTracker — tracks which memories agents actually use vs ignore
• FeedbackAwareScorer — post-scoring re-ranker that improves retrieval over time
• record_feedback() and record_usage() on ALMA core class
• All 7 storage backends updated with feedback persistence
• Absorbs concepts from RuVector (MIT, 3.8k stars) and Beads (MIT, 20.7k stars)

Benchmarks

• LongMemEval R@5 = 0.964 (500 questions, confirmed)
• New Feedback Learning Benchmark (FLB) — proves retrieval improves with usage
• Task Dependency Benchmark scaffold (for Phase 2)
• Colab Pro notebook for GPU-accelerated benchmarking

Documentation

• Updated README with competitor comparison table
• Updated GUIDE.md with feedback loop usage guide
• New Excalidraw architecture diagrams
• Website updated at alma-memory.pages.dev

JS SDK

• alma-memory-js bumped to v0.9.0 with feedback types

Full Changelog: v0.9.0...v0.10.0