Hybrid search across knowledge entries and/or products.
Product results come back in four sections (treffer / ergebnisse / aehnliche / vielleicht) plus the legacy flat lists for backward compatibility with consumers that haven’t migrated yet.
Pipeline (Block C strategy): Pre-Filter → parallel(Exact, BM25, Semantic) → RRF-Fusion → Cross-Encoder Reranker → Section Split → optional Auto-Similar (Centroid-KNN)
Documentation Index
Fetch the complete documentation index at: https://docs.big-panda.ai/llms.txt
Use this file to discover all available pages before exploring further.
API-Token aus dem Big-Panda-Admin-UI unter Einstellungen → API-Keys. Header: Authorization: Bearer <token>.
Space-separated scope set from the token (ADR-015)
Unified search across knowledge entries and/or products.
The sources parameter controls which data pools are searched.
Per-source tuning lives in knowledge (and later products)
sub-objects so consumer-types can scope each pool separately.
1 - 2000Which sources to search: 'knowledge', 'products', or both
1 <= x <= 1000 <= x <= 1Filter by language
Filter products by site (UUID or slug)
Optional per-source configuration for knowledge entries.
Sprint 7.3 — caller can scope the knowledge results without changing the rest of the search request. Empty/None means "no constraint on this axis".
Successful Response
Combined search results.
Two views on the same data:
Plus search_log_id — the row-id of the persisted SearchLog,
so the widget can later POST click events tied to this search.