langgraph-store-firestore

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

pip install langgraph-store-firestore

from langgraph_store_firestore import FirestoreStore

store = FirestoreStore(project_id="my-gcp-project")
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"})

Use it as a LangGraph store: graph.compile(store=FirestoreStore(...)). Async methods work too (sync calls run in a thread). Authentication uses Application Default Credentials (gcloud auth application-default login).

Semantic search (Firestore vector search)

Pass a LangGraph IndexConfig and the store embeds the configured fields on put (stored as a Firestore Vector) and ranks search(query=...) by cosine similarity using find_nearest:

from langgraph_store_core import bedrock_titan_embeddings
from langgraph_store_firestore import FirestoreStore

store = FirestoreStore(project_id="my-gcp-project", collection="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.

Firestore requires a vector index. With create_index=True (default) the store creates a composite index prefix ASC + embedding (flat, dims) on the collection through the Firestore Admin API if one is missing; the build is asynchronous and query searches fail with FAILED_PRECONDITION until it is READY (usually a few minutes; check gcloud firestore indexes composite list). The caller needs datastore.indexes.create / list permissions, or create the index yourself and pass create_index=False:

gcloud firestore indexes composite create --collection-group=memory --query-scope=COLLECTION \
  --field-config=order=ASCENDING,field-path=prefix \
  --field-config=vector-config='{"dimension":"1024","flat":"{}"}',field-path=embedding

The namespace prefix is a range pre-filter on the vector query. Firestore pre-filters are limited to indexed equality/range fields, so value filters are applied on the returned candidates (the store oversamples when a filter is given).

Data model

A single collection of documents {prefix, key, value, created_at, updated_at[, embedding: Vector]}, doc id = enc(prefix)__enc(key). Search is a prefix range query; filters and namespace matching are evaluated in the core.

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

License

MIT