NEURA Knowledge icon

NEURA Knowledge

Search and interact with NEURA enterprise knowledge base

Actions4

RAG Search → Search

AI-generated

Summary

Searches the NEURA knowledge base with hybrid retrieval, combining vector and BM25 search with optional query decomposition and document filters. Returns matching passages with scores and metadata, including Nextcloud links for eligible files.

Inputs

  • Query (required) — Search query for the knowledge base.
  • Top K — Maximum number of results to return; defaults to 20 (UI range: 1–999).
  • Options: BM25 Weight — Hybrid search weighting from 0 (vector only) to 1 (BM25 only); defaults to 0.4.
  • Options: Chat ID — Chat ID for Redis memory management and refinement mode.
  • Options: Conversation History — JSON array of previous queries to provide context; defaults to an empty array. Invalid JSON is ignored.
  • Options: Doc Type Filter — Filter results by document type, such as CONTRATTO or BANDO.
  • Options: Enable Multi-Query — Enable decomposition of complex queries; defaults to true.
  • Options: Include Hierarchy — Include folder hierarchy information; defaults to true.
  • Options: Max Sub Queries — Maximum generated sub-queries when multi-query is enabled; defaults to 3 (UI range: 1–5).
  • Options: Uploaded File IDs — JSON array of OpenWebUI uploaded file IDs to include; defaults to an empty array. Invalid JSON is ignored.

Output shape

A single JSON object per input item, containing query, total_found, retrieval_mode, sub_queries, hierarchy_info, results, and metadata.

Each result includes its index, content, score, filename, path, owner, document type, upload flag, Nextcloud URL (null when unavailable), and metadata. The request is scoped to the authenticated Nextcloud login. Errors stop execution unless Continue On Fail is enabled; then an item with error and itemIndex is returned.

Links

Discussion