Documents — QA harness

Upload PDF/DOCX/XLSX → LLM generates a description → indexed by pgvector + tsvector for hybrid search. Tests the 1C confirm flow and 2C latest+version_count UX.

Base URL: (leave blank to use the page's origin)

0. All documents in knowledge base

1. Upload

2. Search documents (description-level — "which doc?")

2b. Search content (chunk-level — "which passage?")

3. List versions (substring match — case-insensitive)

4. Ask AI — agent answer + cited passages