RAGless Knowledge Base MCPOpen Archive Manager

Archive Manager

No Vectors · No Chunking · No Hallucinations

Leave behind the fragmented chunking and vector drift of traditional RAG. Archive Manager preserves complete document structure via a hierarchical tree index — letting LLMs read and retrieve long documents like an expert, returning precise, traceable paragraph text rather than stitched fragments.

No Vector Database
Full Hierarchical Tree Index
Zero Fragmentation
Fully Transparent Retrieval Path
MCP Plug-and-Play
Chinese-English Mixed Support
Core Difference

Why Not Vector RAG?

Traditional RAG trades fragmentation for scalability — at the cost of systematic precision loss. Archive Manager trades structure for genuine accuracy.

Traditional Vector RAG
  • Documents sliced into fixed-length chunks — context severed
  • Vector similarity matching — semantic drift causes hallucinations
  • Chunks span paragraphs — retrieval results lack completeness
  • Vector database maintenance required — high ops cost
  • Performance degrades sharply for long documents (100+ pages)
  • Retrieval results cannot be traced back to original paragraph structure
Archive Manager RAGless
  • Hierarchical tree index — document structure and chapter relationships fully preserved
  • LLM multi-step reasoning retrieval — no vector drift, zero hallucinations
  • Returns full original paragraphs, accurate to section numbers
  • No vector database dependency — lightweight deployment
  • Optimized for long documents (research reports, regulations, manuals)
  • Full citation path traceable — results verifiable
How the Hierarchical Tree Index Works
① Structure Parsing
  • Document → heading hierarchy detection
  • Build chapter parent-child tree
  • Each node generates content summary
  • Keywords and entity annotation
② Reasoning Retrieval
  • Query → directory tree scan
  • Locate relevant chapter scope
  • Deep-read into nodes
  • Cross-chapter result synthesis
③ Precise Return
  • Complete original paragraph text
  • Precise section path citation
  • Confidence and relevance explanation
  • Retrieval reasoning path auditable

Core Features

Full coverage from knowledge base construction to retrieval to integration

Hierarchical Tree Index Engine

After upload, the document is auto-parsed for heading hierarchy to build a complete directory tree. Each node corresponds to a real document section, preserving parent-child and sibling relationships. LLMs reason over this tree rather than doing similarity matching over fragments.

  • Auto-detects H1-H6 and implicit heading levels
  • Smart structure inference for documents without a table of contents
  • Each tree node includes a summary and keywords for faster targeting

Reasoning-Based Multi-Step Retrieval

The retrieval process mimics how a human expert reads: first scan the table of contents to narrow scope, then deep-read relevant sections, then synthesize across chapters. Every reasoning step is recorded and fully explainable.

  • Two-phase: top-down TOC scan → section deep-read
  • Cross-document multi-source integration with automatic contradiction detection
  • Full retrieval path transparent — every decision auditable

Neo4j Knowledge Graph

Optional module: automatically extracts entities (people, organizations, concepts, events) and relationships from documents to build a cross-document knowledge graph. D3.js force-directed graph for real-time visualization with click-to-explore navigation.

  • Automatic entity and relationship extraction
  • Cross-document entity disambiguation and merging
  • D3 force-directed interactive visualization

Multi-Format Document Support

Comprehensive support for common enterprise document formats, automatically handling complex layouts (tables, captions, footnotes, appendices) to ensure content extraction completeness above 99%.

  • PDF (scanned OCR + native)
  • Word DOCX / Markdown / TXT
  • Auto-processes tables, footnotes, nested lists

MCP Server Standard Interface

Provides MCP-protocol-compliant Server endpoints. Any MCP-compatible AI Agent (including Claude, Lyna Agent, etc.) can mount the knowledge base directly via API Key authentication — no additional integration development required.

  • Standard MCP tool definitions — plug and play
  • API Key fine-grained permission control
  • Supports mounting multiple knowledge base instances simultaneously

Multi-Scenario Preset Configurations

Built-in retrieval strategy configurations deeply optimized for different document types — achieving optimal retrieval performance in each scenario without manual tuning.

  • General / Legal / Financial / Policy / Technical docs
  • Custom scenario configuration support
  • Scenario configs exportable and reusable

Get Started in 4 Steps

From document upload to Agent-ready — as fast as 5 minutes

01

Upload Documents

Drag and drop PDFs, DOCX, or Markdown — supports bulk upload of entire folders.

02

Auto Index

The system parses document structure, builds a hierarchical directory tree, and generates node summaries and keyword indexes.

03

Natural Language Search

Ask questions directly in plain language. The system reasons over the knowledge tree and returns precise original paragraphs.

04

Agent Integration

Mount to any AI Agent via MCP interface to enable knowledge-driven automated workflows.

Technical Advantages

From architectural design to engineering implementation — every detail serves precise retrieval

No Vector Database

Completely free from vector database ops burden. Knowledge indexes are stored as structured tree data — no Pinecone, Weaviate, or Chroma infrastructure needed. Simple deployment, low cost.

Incremental Document Updates

When documents are added or modified, the system only rebuilds changed nodes rather than the full index. Large knowledge base (thousands of documents) update time drops from minutes to seconds.

Paragraph-Level Precision

Retrieval results are accurate to paragraph level with complete citation paths (document → section → sub-node number). Citation accuracy approaches 100%, meeting strict compliance requirements for legal and financial use cases.

Quantifiable Retrieval Quality

Each retrieval returns a confidence score and relevance explanation. The system records retrieval history and user feedback to continuously optimize retrieval strategies for similar documents.

Multilingual Documents

Native support for Chinese-English mixed documents. Auto-detects the primary document language, intelligently translates search terms for cross-language queries, and manages Chinese and English knowledge bases together.

Async Indexing, Instant Availability

Documents can be queried immediately after upload — no need to wait for full indexing to complete. Indexing a large document (500-page PDF) typically completes within 30 seconds.

MCP Protocol Integration

Give Any AI Agent Knowledge Base Capabilities

Archive Manager provides a standard MCP Server interface. Any MCP-compatible AI Agent can mount the knowledge base through configuration — no integration code required.

API Key authentication with fine-grained permission control
One Agent can mount multiple knowledge bases simultaneously
Query latency typically under 2 seconds
Complete call logs and usage statistics
mcp_config.json
{
  "mcpServers": {
    "archive-manager": {
      "url": "https://archive.runemind.com.cn/mcp",
      "apiKey": "am_sk_xxxxxxxxxxxxxxxx",
      "knowledgeBases": [
        "legal-docs",
        "product-manual",
        "research-reports"
      ]
    }
  }
}
Configuration complete — Agent can now query the knowledge base directly

Use Cases

Any scenario requiring precise information retrieval from long documents

Enterprise Knowledge Base

Product ManualsInternal StandardsTechnical Docs

Consolidate product manuals, internal standards, technical documentation, and meeting minutes into one KB. New employees ask onboarding questions, veterans look up policies — all through natural language, no manual FAQ maintenance required.

Legal & Compliance Research

Contract AnalysisRegulation SearchCase Research

Upload contracts, regulations, and court rulings. AI precisely locates relevant clauses with full contextual understanding — no out-of-context misinterpretation from chunking. Ideal for law firms and compliance departments.

Financial Report Analysis

Research ReportsAnnual ReportsProspectus Review

Cross-document retrieval and comparative analysis across large volumes of research reports, annual reports, and prospectuses. Organize knowledge trees by company, industry, and time dimension to quickly locate specific financial data or analytical views.

Technical Documentation Assistant

API DocsArchitecture DesignOps Manuals

Consolidate API docs, architecture design documents, and operations manuals. Developers and ops teams ask questions in natural language and get precise answers — no more Ctrl+F through piles of documents.

Start Building Your Knowledge Base

Upload documents and get precise, traceable knowledge retrieval within 5 minutes. No vector database, no complex setup — ready out of the box.