Knowledge Management Skill

Slice: slices/knowledge/ Type: Knowledge Base Operations

Purpose

Hi-RAG v2 knowledge base management. Use this skill when:

• •Searching the knowledge base for information
• •Ingesting new documents into the knowledge base
• •Managing knowledge index health

Quick Start

from slices.knowledge import KnowledgeService, KnowledgeTask, RetrievalMode

service = KnowledgeService(http_client=client)

# Query knowledge
task = KnowledgeTask(
    query="How does TensorZero routing work?",
    retrieval_mode=RetrievalMode.HYBRID,
    top_k=10,
    rerank=True,
)
result = await service.execute(task)

for chunk in result.chunks:
    print(f"[{chunk.score:.2f}] {chunk.content[:100]}")

API Reference

KnowledgeService

Retrieval Modes

Index Operations

Integration Points

Example: Document Ingestion

from slices.knowledge import KnowledgeTask, IndexOperation

# Ingest new documents
task = KnowledgeTask(
    operation=IndexOperation.INGEST,
    documents=[
        {
            "content": "TensorZero is a unified LLM gateway...",
            "source": "docs/tensorzero.md",
            "metadata": {"type": "documentation"},
        }
    ],
)
result = await service.execute(task)
print(f"Processed {result.documents_processed} documents")

Retrieval Strategy

1. •Hybrid Mode (Default): Combines all three retrieval methods
2. •Cross-encoder Reranking: Improves relevance with neural reranker
3. •Filtering: Apply metadata filters to narrow results
4. •Top-K: Control number of results returned

Best Practices

• •Use HYBRID mode for general queries
• •Use VECTOR for conceptual/semantic search
• •Use FULLTEXT for exact phrase matching
• •Use GRAPH for entity relationship queries
• •Always enable reranking for best relevance