RAGFlow vs Dify KBs — head-to-head on one complex PDF

Test material#

A real electronics manual PDF:

• 30 pages
• 12 tables (specs, comparisons, error codes)
• 6 pages of two-column layout
• 2 scanned pages (appendix from older edition)

Test questions (30)#

• “What’s the max output power of model X-200?” (table)
• “How do I handle error code E04?” (table)
• “What’s the temperature limit mentioned in the two-column section?” (layout)
• “What safety rules are in appendix B (scanned)?” (OCR)

Results#

Metric	Dify default	Dify tuned	RAGFlow default	RAGFlow tuned
Tables top-1	6/12	9/12	11/12	12/12
Two-column top-1	2/6	4/6	6/6	6/6
Scanned top-1	0/2	0/2	2/2	2/2
Plain paragraphs top-1	9/10	10/10	9/10	10/10
Overall	17/30	23/30	28/30	30/30

Why RAGFlow lands the haymaker on complex docs#

1. DeepDoc parsing identifies title / paragraph / table / figure layouts before chunking
2. Tables preserve structure — header + cells stay 2D; an entire table is one chunk
3. Built-in OCR for scans
4. Multi-route recall — vector + BM25 + knowledge graph, fused

Why Dify is still fine for plain paragraphs#

Dify’s strengths are API consistency and ecosystem. If your KB is 90% Markdown / Notion exports without layout, Dify suffices and is lighter.

Picking#

Your docs are mostly	Pick
Markdown / Notion / web	Dify
API reference / simple PDFs	Dify
Product manuals with many tables	RAGFlow
Scanned / legacy docs	RAGFlow
Mixed, max-quality goal	RAGFlow as retrieval backend + Dify as app layer

Combo architecture (production-recommended)#

Dify governs workflow + multi-LLM. RAGFlow does what it does best — complex parsing + multi-route recall.