Vectorstore Backend Base Class

The interface of vector database backends.

A backend runs one vectorstore’s database operations: its lifecycle, its data, and its dumps. Lifecycle, status and scheduling are the job of VectorstoreService, which calls a backend.

class smarter.apps.vectorstore.backends.base.SearchResult(id, text, metadata=<factory>, score=None)

Bases: object

A chunk that a search found.

__init__(id, text, metadata=<factory>, score=None)

id: str | None

metadata: dict[str, Any]

score: float | None = None

text: str

class smarter.apps.vectorstore.backends.base.SmarterVectorstoreBackend(vectorstore, embeddings=None)

Bases: ABC

Abstract base class of the vector database backends.

Parameters:

• vectorstore (VectorstoreMeta) – The vectorstore.

• embeddings (Embeddings | None) – Its embeddings model. Required for upsert() and search().

__init__(vectorstore, embeddings=None)

abstractmethod create()

Create the index or collection.

Return type:

None

abstractmethod create_snapshot(name)

Take a snapshot, or backup.

Return type:

SnapshotInfo

abstractmethod delete(ids)

Delete chunks.

Ids that do not exist are ignored.

Return type:

None

abstractmethod delete_snapshot(name)

Delete a snapshot, or backup.

Return type:

None

destroy()

Delete the index or collection, and any infrastructure that Smarter runs.

Return type:

None

abstractmethod drop()

Delete the index or collection, and its data.

Return type:

None

property embeddings: Embeddings

abstractmethod exists()

Whether the index or collection exists.

Return type:

bool

property index_name: str

infrastructure_ready()

Whether the infrastructure that Smarter runs is ready, and why not.

A managed service always is.

Return type:

tuple[bool, str]

logs(tail=200)

The database server’s logs, if Smarter runs it.

Return type:

str | None

provision()

Create the database’s infrastructure, if Smarter runs it.

It returns without waiting for it.

Return type:

None

abstractmethod restore_snapshot(name)

Replace the index or collection’s data with a snapshot’s.

Return type:

None

abstractmethod search(query, k=4, search_type='similarity', score_threshold=None, fetch_k=None, lambda_mult=None, metadata_filter=None)

Find the chunks that are closest in meaning to a query.

Parameters:

• search_type (str) – similarity, similarity_score_threshold, or mmr (maximal marginal relevance).

• metadata_filter (dict[str, Any] | None) – chunks whose metadata has these values, e.g. {“source”: “faq”}.

Return type:

list[SearchResult]

abstractmethod stats()

The database’s statistics.

It includes vector_count.

Return type:

dict[str, Any]

stop()

Stop the infrastructure that Smarter runs, keeping its data.

Return type:

None

abstractmethod upsert(ids, texts, metadatas, batch_size=64)

Embed and insert, or replace, chunks.

Return type:

None

class smarter.apps.vectorstore.backends.base.SnapshotInfo(name, size_bytes=None, ready=True)

Bases: object

A Qdrant snapshot, or a Pinecone backup.

__init__(name, size_bytes=None, ready=True)

name: str

ready: bool = True

size_bytes: int | None = None

exception smarter.apps.vectorstore.backends.base.VectorStoreBackendConnectionError(message='')

Bases: VectorStoreBackendError

The vector database cannot be reached.

exception smarter.apps.vectorstore.backends.base.VectorStoreBackendError(message='')

Bases: SmarterException

A vector database operation failed.