Semantic similarity
A measure of how closely related two pieces of text are in meaning, computed as the distance (typically cosine similarity) between their embeddings in vector space. The primary signal for k-NN retrieval in the memory gateway. High semantic similarity does not require lexical overlap.
Semantic similarity is what allows the memory gateway to retrieve memories about 'service latency increasing under load' when the query is about 'response time degradation at high throughput' - the meanings are closely related even though the exact words differ. It is computed by encoding both the query and stored memory summaries into embedding vectors and measuring the cosine angle between them. Vectors pointing in similar directions (small cosine angle, high cosine similarity) represent semantically close content. The quality of semantic similarity search depends heavily on the embedding model: a model that produces well-separated embeddings for distinct concepts and closely-clustered embeddings for related ones will support more useful retrieval than one that distributes embeddings more uniformly. The multi-lane embedding architecture (quality, efficient, hybrid profiles) allows the memory gateway to trade accuracy for speed depending on current budget constraints.