Skip to content

knowledge_search

KnowledgeSearchTool — filtered BM25 retrieval with source attribution.

Wraps KnowledgeStore so agents can search ingested documents by text query and optional provenance filters (source, doc_type, author, date range). Optionally delegates to a TwoStageRetriever for BM25 + reranking.

Classes

KnowledgeSearchTool 

KnowledgeSearchTool(store: Optional[KnowledgeStore] = None, retriever: Optional['TwoStageRetriever'] = None)

Bases: BaseTool

Search the knowledge store using filtered BM25 retrieval.

Results include source attribution so agents can cite provenance. When a TwoStageRetriever is supplied it is used in place of the store's direct retrieve method, enabling optional semantic reranking.

Source code in src/openjarvis/tools/knowledge_search.py 
32
33
34
35
36
37
38
def __init__(
    self,
    store: Optional[KnowledgeStore] = None,
    retriever: Optional["TwoStageRetriever"] = None,
) -> None:
    self._store = store
    self._retriever = retriever