langgraph-store-postgres

A PostgreSQL long-term-memory store (BaseStore) for LangGraph — namespaced key/value memory with prefix search, filters, list_namespaces, and native semantic search via pgvector.

pip install langgraph-store-postgres

from langgraph_store_postgres import PostgresStore

store = PostgresStore("postgresql://user:pass@localhost:5432/db")   # or engine=<Engine>

store.put(("users", "1", "memories"), "food", {"text": "loves sushi", "kind": "pref"})
item = store.get(("users", "1", "memories"), "food")
hits = store.search(("users", "1"), filter={"kind": "pref"}, limit=10)
spaces = store.list_namespaces(prefix=("users",))

Use it as a LangGraph store: graph.compile(store=PostgresStore(...)). Async methods (aget/aput/asearch/…) work too — the sync calls run in a thread.

Semantic search (pgvector)

Pass a LangGraph IndexConfig and the store embeds the configured fields on put and ranks search(query=...) by cosine similarity using pgvector (<=>, HNSW index):

from langgraph_store_core import bedrock_titan_embeddings
from langgraph_store_postgres import PostgresStore

store = PostgresStore(url, table_name="memory",
                      index={"dims": 1024, "embed": bedrock_titan_embeddings(dimensions=1024), "fields": ["text"]})
store.put(("memories", "kamal"), "k1", {"text": "the user loves sushi", "kind": "pref"})
hits = store.search(("memories", "kamal"), query="what food does the user like?", filter={"kind": "pref"})
print(hits[0].score, hits[0].value)

embed may be any LangChain Embeddings, a list[str] -> list[list[float]] callable, or a provider string. fields defaults to ["$"] (whole value as JSON). put(..., index=False) skips embedding for one item.

Requires the pgvector extension on the server (apt install postgresql-16-pgvector, or built in on RDS / Cloud SQL / Azure Database). The store runs CREATE EXTENSION IF NOT EXISTS vector, adds an embedding vector(dims) column and an HNSW cosine index. Namespace prefix and plain-equality filters run inside the SQL query (JSONB containment); operator filters ($gt, $in, …) are applied on the returned candidates.

Data model

A single table (prefix, key, value JSONB, created_at, updated_at[, embedding vector]) with primary key (prefix, key), created automatically. Namespace tuples are joined with a unit separator into prefix; search is a prefix scan, filters and namespace matching are evaluated in the core.

Docs: https://skamalj.github.io/agentstate-reducer/

License

MIT