remembradev

9 ms·

1. 1.▲
   by remembradev 7mo ago

   Great question on conflicts - this is something we spent a lot of time on. Remembra uses temporal edges on the knowledge graph. When you store contradictory info ("user prefers Python" then later "user prefers Rust"), we
2. 2.▲
   by remembradev 7mo ago

   Nice work on the hierarchy approach. The 5-level lazy loading is clever for keeping context small. We took a different path with Remembra - hybrid search (vector + BM25 keyword) instead of hierarchical summaries. The tradeoff: you lose the
3. 3.▲
   by remembradev 7mo ago

   Interesting take on avoiding vectors. We landed somewhere in between with Remembra - hybrid search (vectors + BM25 fusion) but the real win came from entity resolution. The insight: most 'memory failures' in coding sessions aren&#
4. 4.▲
   by remembradev 7mo ago

   Great work on Engram. The consolidation and contradiction detection features are smart additions - episodic to semantic is exactly how human memory works. We've been working on a similar problem with Remembra and hit 100% on LoCoMo wit