UNDERSTAND
LibRAG precise content recall engine
Precise retrieval.
Ready on day one.
Recall and precision both exceed 95%. No endless chunking, threshold, model, or prompt tuning. Import documents, build a semantic hierarchy automatically, and answer from reliable evidence on day one.
Policies · Financial reports · Tender documents · Enterprise knowledge · Private deployment
01 · ZERO TUNING
A useful RAG system
should not begin with months of tuning.
Traditional RAG often requires repeated adjustments to chunking, embeddings, thresholds, Top K, reranking, and prompts. When the data changes, the tuning cycle begins again.
LibRAG absorbs that complexity into the engine: import documents, build the index automatically, and start using it.
- 01
Automatic indexing
Semantic chunking and hierarchical classification are completed during ingestion without per-library parameter hunting.
- 02
Automatic routing
The engine understands intent, narrows the search level by level, then locates the evidence needed for the answer.
- 03
Automatic verification
Built-in evaluation keeps retrieval, generation, and release changes observable instead of relying on ad-hoc questions.
02 · CONTEXT ENGINE
Archival science.
Beyond similarity.
LibRAG brings classification, hierarchy, and provenance into context engineering to create a controllable, explainable, and reasoning-ready answer context for large language models.
LOCATE
Understand where knowledge lives
Move through domain, category, document, and passage to handle scenarios, multi-hop reasoning, and comparisons.VERIFY
Verify the basis of the answer
Rerank the evidence that truly matters and preserve sources so every conclusion can return to the original text.03 · BUILT-IN RAG EVAL
Evaluation is not an add-on.
It closes the quality loop.
From standardized cases and automated runs to LLM Judge and failure attribution, LibRAG turns retrieval quality into an engineering process that is measurable, repeatable, and explainable.
- Case setsLock questions, gold evidence, and acceptance criteria
- Batch runsEvaluate retrieval and generation at scale
- DiagnosisSeparate recall gaps, generation drift, and citation issues
- RegressionCompare the actual impact of model, prompt, and knowledge changes
04 · RAGAS POC
One yardstick.
Results you can inspect.
1,000 aligned questions cover direct retrieval, multi-hop QA, reasoning retrieval, and content comparison. All four systems were evaluated with the same Ragas toolkit and Qwen judge across context precision, context recall, faithfulness, and answer relevancy.
LibRAG leads three of the four metrics. Other vendor C leads faithfulness; the result is shown as measured.
Context precision 0.9779BEST
Context recall 1.0000BEST
Faithfulness 0.7514SECOND
Answer relevancy 0.8982BEST
| System | Context precision | Context recall | Faithfulness | Answer relevancy |
|---|---|---|---|---|
| LibRAG | 0.9779 | 1.0000 | 0.7514 | 0.8982 |
| Other vendor A | 0.95 | 0.92 | 0.45 | 0.87 |
| Other vendor B | 0.80 | 0.95 | 0.70 | 0.83 |
| Other vendor C | 0.52 | 0.85 | 0.79 | 0.69 |
05 · BUSINESS READY
One retrieval foundation for demanding business workflows.
Go beyond factual lookup to handle scenarios, multi-hop questions, content comparisons, and evidence verification.
- 01
Enterprise Knowledge QA
Search PDF, Word, images, webpages, and tables with answers that link back to source evidence.
- 02
Compliance assistant
Locate policies, regulatory documents, and case evidence for compliance QA and review workflows.
- 03
Report intelligence
Understand financial reports and table structures, then locate, compare, and explain key data across documents.
- 04
RAG Eval
Use standard cases, automated runs, and evidence chains for launch acceptance and regression testing.
- 05
Ingestion enhancement
Transform raw files into high-quality traceable passages that integrate with an existing RAG pipeline.
- 06
Agents and external data
Connect governed enterprise APIs and live data so retrieval becomes a reliable entry point for business intelligence.
06 · PRIVATE DEPLOYMENT
Bring the engine inside.
Keep the data there.
Applications, models, data, and infrastructure can all stay within the enterprise boundary. Connect Web, Agents, Dify, RagFlow, ChatFlow, and HTTP APIs through one service layer.
Access control, logs, provenance, and evaluation follow the same retrieval chain.
BUILT BY YUMBEN
广州云本开源软件有限公司
Guangzhou Yumben Open Source Software Co., Ltd. was founded in 2015 and focuses on enterprise open-source software and technical services for telecommunications, finance, manufacturing, and government. LibRAG is grounded in long-term enterprise delivery experience, not a one-off concept demo.
service@yumben.com07 · EXPERIENCE
Validate “ready out of the box” with real questions.
Explore the core workflow in the public environment, or request an enterprise trial with your own documents and acceptance criteria.
PUBLIC TEST ENVIRONMENT
LibRAG Product Service
Explore knowledge-base, retrieval, answer, and evaluation workflows. Do not upload sensitive or restricted information.
- Test account
demo- Test password
demo123456
Service provider: 广州云本开源软件有限公司