Embedding
A vector representation of text or structured input used for similarity search and retrieval in the memory substrate. The system supports multiple embedding profiles simultaneously: quality (higher accuracy), efficient (lower latency), and hybrid (blend of both).
An embedding is a dense numerical vector that encodes the semantic content of a piece of text. Two pieces of text with similar meanings will have embedding vectors that point in similar directions in the high-dimensional space the model has learned to use. This property - semantic similarity as geometric proximity - is what makes vector search possible: rather than matching keywords, the memory gateway computes an embedding for the query and finds stored memories whose embeddings are geometrically close. The Cognitive Substrate maintains three embedding lanes per experience event simultaneously. The quality lane (Qwen model, 1024 dimensions) prioritizes accuracy and is used when retrieval precision matters most. The efficient lane (Nomic model, 768 dimensions) prioritizes low latency and is used under budget pressure. The hybrid lane (BGE-M3 model, 1024 dimensions) is a general-purpose option that balances both. Keeping all three lanes in the index allows the memory gateway to select the retrieval profile appropriate to the current budget constraint without re-embedding memories on demand.