Project
Mem-on-the-Go
Universal agent memory system built on RuVector. 9 reinforcement learning algorithms, hybrid BM25+vector search with Reciprocal Rank Fusion, GNN 8-head attention (+12.4% recall), and 4 auto-selecting backends. Runs anywhere with zero setup. 61 microsecond search latency with Rust+SIMD acceleration. Self-learning improves retrieval quality 36% over time without manual tuning.
Pipeline Architecture
Click any node to explore details.
Universal agent memory system built on RuVector. 9 reinforcement learning algorithms, hybrid BM25+vector search with Reciprocal Rank Fusion, GNN 8-head attention (+12.4% recall), and 4 auto-selecting backends. Runs anywhere with zero setup. 61 microsecond search latency with Rust+SIMD acceleration. Self-learning improves retrieval quality 36% over time without manual tuning.