Universal High-Performance RAG Knowledge Base for AI Agents
Index local files, websites, GitHub repos, npm/PyPI packages — then search them with hybrid AI-powered retrieval. All through your IDE chat.
Features •
Architecture •
Quickstart •
Tools •
FAQ
✨ Features
🔍 Search Pipeline
Universal Local RAG Server for Model Context Protocol (MCP)
File Watcher — auto-reindex on filesystem changes (watchdog, 2s debounce)
Admin tools — list, remove, delete, reindex — all through chat
100% Local & Free — no API keys, no Docker, no monthly fees
🏗️ Architecture
flowchart TD
Q(["🔍 User Query"])
Q --> QE
subgraph EXPAND["📝 Stage 0 — Query Expansion"]
QE["RU→EN Synonyms<br>(up to 3 variations)"]
end
QE --> VS
QE --> BM
subgraph HYBRID["⚡ Stage 1 — Hybrid Retrieval"]
VS["🧠 Vector Search<br>(ChromaDB + bge-m3)"]
BM["📝 Keyword Search<br>(BM25 Okapi)"]
end
VS --> RRF
BM --> RRF
subgraph FUSION["🔀 Stage 2 — Fusion"]
RRF["Reciprocal Rank Fusion<br>(k=60)"]
end
RRF --> CE
subgraph RERANK["🎯 Stage 3 — Reranking"]
CE["Cross-Encoder<br>bge-reranker-v2-m3"]
end
CE --> DD
subgraph DEDUP["🧹 Stage 4 — Dedup"]
DD["Remove >80%<br>similar results"]
end
DD --> OUT
subgraph OUTPUT["📋 Stage 5 — Result"]
OUT["Top-N Documents<br>with Breadcrumbs"]
end
style Q fill:#6366f1,color:#fff,stroke:#4338ca
style EXPAND fill:#0f172a,color:#e2e8f0,stroke:#334155
style HYBRID fill:#0f172a,color:#e2e8f0,stroke:#334155
style FUSION fill:#0f172a,color:#e2e8f0,stroke:#334155
style RERANK fill:#0f172a,color:#e2e8f0,stroke:#334155
style DEDUP fill:#0f172a,color:#e2e8f0,stroke:#334155
style OUTPUT fill:#0f172a,color:#e2e8f0,stroke:#334155
style QE fill:#581c87,color:#e9d5ff,stroke:#9333ea
style VS fill:#1e40af,color:#bfdbfe,stroke:#3b82f6
style BM fill:#065f46,color:#a7f3d0,stroke:#10b981
style RRF fill:#7c3aed,color:#ede9fe,stroke:#8b5cf6
style CE fill:#b45309,color:#fef3c7,stroke:#f59e0b
style DD fill:#0e7490,color:#cffafe,stroke:#06b6d4
style OUT fill:#1e3a5f,color:#bae6fd,stroke:#38bdf8
How It Works
Query Expansion — generates up to 3 query variations with RU→EN programming synonyms
Hybrid Retrieval — searches by semantic meaning (bge-m3, 8192 tokens) and exact keywords (BM25) simultaneously
Reciprocal Rank Fusion — mathematically combines ranks from both engines
Cross-Encoder Reranking — bge-reranker-v2-m3 deeply computes relevance for top candidates
Deduplication — removes near-identical results
Structured Output — results returned with breadcrumbs (e.g., README.md > Quickstart > Installation)
🚀 Quickstart
1. Prerequisites
Python 3.10+ (Tested on 3.13)
git
2. Install
git clone https://github.com/ElvinBayramov/OmniDocs-RAG.git
cd OmniDocs-RAG
python install.py
The installer downloads AI models (~2.2GB for bge-m3 + bge-reranker), installs dependencies, and auto-configures your IDE.
⚠️ Enable GPU Acceleration (Crucial for Speed):
By default, pip may install the CPU-only version of PyTorch on Windows. To unlock your NVIDIA GPU, run:
> Index my project docs
> Search: how does authentication work?
> Index the FastAPI documentation from https://fastapi.tiangolo.com
> Add the langchain repo: github://langchain-ai/langchain/docs
The AI calls the MCP tools automatically — no UI, no buttons, just chat.
🛠️ MCP Tools
Tool
Description
index_documents(path, collection)
Index local files (40+ formats, incremental)
index_url(uri, collection, ...)
Index websites, GitHub, npm, PyPI, ZIP
search_docs(query, n, category, filename, collection)
Code (wrapped in markdown):.py.js.ts.jsx.tsx.css.java.go.rs.c.cpp.rb.php.swift.kt.lua.sh
Config:.json.yaml.yml.toml.xml.csv.ini.cfg
Binary (optional packages):
Format
Install
PDF
pip install pypdf
Word (.docx)
pip install python-docx
Excel (.xlsx)
pip install openpyxl
PowerPoint (.pptx)
pip install python-pptx
Jupyter (.ipynb)
built-in
🏷️ Auto-Categorization
Every file gets a category automatically (no manual tagging needed):
Priority
Source
Example
1
YAML frontmatter category:
category: architecture → architecture
2
First # Heading in the file
# API Reference → api reference
3
Filename stem
system_design.md → system design
❓ FAQ
Q: Does this send my data anywhere?
A: No. 100% local. Models download once from HuggingFace, then everything runs offline. No API keys, no cloud.
Q: Do I need a GPU?
A: No, but it helps. CPU works fine for search (~200ms). GPU (CUDA) accelerates indexing ~11x. Set RAG_DEVICE=cuda.
Q: How do I update the index?
A: The server uses incremental indexing — only changed files are re-indexed. Just call index_documents() again, or enable the file watcher with RAG_WATCH_PATH.
Q: Why is the first search slow?
A: The Cross-Encoder (~1.1GB) loads lazily on first query. All subsequent searches are instant.
Q: Does it support my language?
A: Yes. bge-m3 supports 100+ languages including English, Russian, Chinese, Japanese, Spanish, and more. Cross-language search works too (query in Russian, find results in English).
Q: Can I have separate knowledge bases per project?
A: Yes. Use the collection parameter: index_documents(path, collection="my-project"), then search_docs(query, collection="my-project").
📄 License
Licensed under the Apache License 2.0. See LICENSE for details.
O
OmniDocs RAG
ElvinBayramov·4 Sterne·
Universal Local RAG Server for Model Context Protocol (MCP)
Noch keine Tool-Liste. Remote-Server erhalten eine nach dem ersten Health-Check.
FAQ
Fragen
Wie installiere ich OmniDocs RAG?
Kopieren Sie auf dieser Seite die stdio-Konfiguration für Cursor, Claude Desktop, Claude Code oder VS Code. Sie nutzt die Installation des Herausgebers, keinen PlugMCP-Proxy.
Benötigt OmniDocs RAG einen API-Schlüssel?
In diesem Eintrag ist keine erforderliche Umgebungsvariable aufgeführt.
Welchen Transport verwendet OmniDocs RAG?
OmniDocs RAG läuft lokal über stdio.
Wann wurde OmniDocs RAG zuletzt geprüft?
Zuletzt geprüft: 2026-10-28. Universal Local RAG Server for Model Context Protocol (MCP)