Simple Knowledge Base

A simple, local-first semantic knowledge base.

Add your documents. Search with natural language. Find what matters.

Simple Knowledge Base uses semantic chunking and vector embeddings to understand your documents, not just match keywords. Results are reranked for relevance, so the best answers surface first.

✨ Features

• Multi-Index Support — Organize documents into separate searchable collections
• Semantic Search — Find information by meaning, not just keywords
• Smart Chunking — Documents are split at natural boundaries for better context
• Reranked Results — Cross-encoder reranking puts the best matches first
• Local & Private — Everything runs on your machine, no cloud required
• Simple API — Clean REST endpoints for easy integration
• MCP Server — Integrate with AI assistants via Model Context Protocol

📸 Screenshots

Search Knowledge Base

Search through your documents using natural language queries. Results are ranked by semantic relevance with visual indicators showing match quality.

Add Knowledge

Create indexes to organize your documents and add content through single file upload or batch directory processing.

Quick Start

# Install all dependencies
make install

# Start both frontend and backend
make dev

This launches:

• Backend: http://localhost:8000 (FastAPI + LanceDB)
• Frontend: http://localhost:5173 (React + Vite)
• API Docs: http://localhost:8000/docs (Swagger UI)

Press Ctrl+C to stop both services.

Development Commands

make help        # Show all available commands
make dev         # Start both frontend and backend
make backend     # Start only backend
make frontend    # Start only frontend
make install     # Install all dependencies
make test        # Run backend tests
make lint        # Run linters
make clean       # Clean generated files
make mcp-build   # Build the MCP server
make mcp-dev     # Run MCP server in dev mode
make mcp-inspect # Test MCP server with Inspector

Project Structure

simple-knowledge-base/
├── backend/          # FastAPI backend (Python 3.13)
│   ├── app/          # Application code
│   ├── tests/        # Test suite
│   └── README.md     # Backend documentation
├── frontend/         # React frontend (Vite + TypeScript)
├── mcp-server/       # MCP server for AI assistant integration
│   ├── src/          # TypeScript source
│   └── README.md     # MCP server documentation
├── scripts/          # Development scripts
│   ├── dev.sh        # Combined dev server
│   └── start-backend.sh
├── .github/images/   # Screenshots and assets
└── Makefile          # Development commands

Tech Stack

Layer	Technology
Frontend	React 19, Vite, Cloudscape Design
Backend	FastAPI, Python 3.13
Vector DB	LanceDB
Embeddings	Alibaba-NLP/gte-multilingual-base
Reranker	Alibaba-NLP/gte-multilingual-reranker-base
MCP Server	TypeScript, MCP SDK

MCP Server

The MCP (Model Context Protocol) server allows AI assistants to search your knowledge base directly.

Quick Setup

# Build the MCP server
cd mcp-server
npm install
npm run build

Configure Your MCP Client

Add to your MCP configuration (e.g., .kiro/settings/mcp.json):

{
  "mcpServers": {
    "simple-kb": {
      "command": "node",
      "args": ["/path/to/simple-knowledge-base/mcp-server/dist/index.js"],
      "env": {
        "KB_API_URL": "http://localhost:8000"
      }
    }
  }
}

Available Tools

Tool	Description
kb_list_indexes	List all available indexes
kb_search_docs	Semantic search with ranked results
kb_fetch_doc	Fetch chunks from a specific document

See mcp-server/README.md for detailed documentation.

Documentation

• Backend API Documentation
• MCP Server Documentation

License

This project is licensed under the MIT License - see the LICENSE file for details.