pgvector — same Postgres you already run

-- CREATE EXTENSION vector;
CREATE TABLE adk_memory (
  id uuid PRIMARY KEY,
  user_id text NOT NULL,
  content text NOT NULL,
  embedding vector(768) NOT NULL,
  created_at timestamptz DEFAULT now()
);
CREATE INDEX ON adk_memory USING hnsw (embedding vector_cosine_ops);
Advertisement

Vertex Matching Engine — managed

Google's managed vector store. Scales to billions of vectors. More expensive but zero ops. Use for cross-cluster shared memory.

Advertisement

Self-hosted Qdrant / Weaviate

Both open source, both scale well. Qdrant for simplicity, Weaviate for hybrid keyword + vector search.