Steve HutchinsonBig Pines

Vector search

Retrieval based on embedding similarity rather than keyword matching. Finds memories that are semantically related to a query even when they share no exact terms. Implemented in OpenSearch using HNSW indexes for approximate nearest neighbor lookup. The k-NN component of the memory gateway's hybrid retrieval pipeline.

Vector search is what enables the memory gateway to surface contextually relevant memories that do not share keywords with the query. A query about 'latency spikes under load' will surface memories about 'response time degradation during peak traffic' because their embedding vectors are close in the high-dimensional space the model has learned to represent semantic relationships. The core infrastructure is OpenSearch's k-NN plugin, which uses HNSW indexing to make approximate nearest neighbor search fast at scale. The memory substrate maintains multiple embedding lanes simultaneously - quality (1024-dimensional Qwen embeddings), efficient (768-dimensional Nomic embeddings), and hybrid (1024-dimensional BGE-M3 embeddings) - allowing the memory gateway to select the retrieval profile that best matches the current budget constraints. Full semantic similarity requires the quality lane; fast retrieval under budget pressure uses the efficient lane.

This site collects anonymous usage data to understand how people read and navigate the blog. Accepting enables persistent reader preferences across visits.