vectorize-search

Semantic search over encrypted memories is implemented by indexing plaintext content and storing generated embeddings with record IDs in a vector index. A workflow then decrypts the retrieved results after the search is performed.

Vectorize enables agent memory retrieval by upserting and querying vector indexes using metadata filters. This allows targeted recall of relevant memories by querying the index with natural language across a federated network.

Yes, an embedding model is required to embed text into dense representations for similarity search. You also need a Vectorize index and a workflow to handle the encrypted memory retrieval process.

You can retrieve relevant memories by running a semantic search against a natural language query. The system matches the query against dense vector representations stored in the index to return matching records.

You should avoid using a vector database for memory recall if your workflow cannot decrypt results after retrieval or if you lack an embedding model to generate the dense representations required for indexing plaintext content.