plexus.analysis.s3vectors_client module

S3 Vectors client for Semantic Reinforcement Memory.

Thin adapter around boto3 s3vectors operations with helper methods used by the VectorTopicMemory report block.

class plexus.analysis.s3vectors_client.TopicMemoryVectorStore(bucket_name: str, index_name: str, region: str, index_arn: str | None = None, client=None)

Bases: object

S3 Vectors-backed topic memory store.

Supports health check, full rebuild, cluster persistence, and cluster metadata updates used by Semantic Reinforcement Memory.

__init__(bucket_name: str, index_name: str, region: str, index_arn: str | None = None, client=None)

build_index(documents: List[Dict[str, Any]], old_index_for_swap: str | None = None) → Dict[str, Any]

Rebuild index contents by clearing existing vectors and writing new items.

old_index_for_swap is accepted for API compatibility with the previous vector store adapter and intentionally ignored.

bulk_update_cluster_weights(clusters: List[Dict[str, Any]], index_name: str | None = None) → None

Update memory_weight and memory_tier on existing cluster vectors.

clear_index(index_name: str | None = None) → int

Delete all vectors in the index. Returns number of deleted vectors.

find_nearest_clusters(embedding: ndarray, k: int = 5, threshold: float | None = None) → List[Dict[str, Any]]

Query nearest cluster vectors.

Returns results ordered by descending similarity with compatibility fields. threshold is interpreted as a minimum similarity score, matching prior vector store adapter behavior.

get_all_clusters() → List[Dict[str, Any]]

Retrieve all cluster vectors from the index.

health_check() → bool

Return True if the configured vector index is reachable.

persist_clusters(clusters: List[Dict[str, Any]], index_name: str | None = None) → None

Bulk-persist cluster vectors and metadata.