KuraDB

[!NOTE] This README was generated by SKILL, get the ZH version from here.

DROP FILES IN, LET YOUR AGENT SEARCH THEM OUT

A Go read-only RAG database with drop-in auto-indexing, parallel keyword and semantic search, and MCP tools

Table of Contents

• Features
• Architecture
• License
• Author

Features

git clone https://github.com/pardnchiu/KuraDB.git && cd KuraDB && make app · Documentation

• Drop-In Indexing — Drop files into ~/Kura_{name} and the watcher parses PDF, DOCX, PPTX, CSV, XLSX, and plain text, queues them for embedding, and soft-deletes removed files.
• Parallel Keyword + Semantic Search — Each request runs gse-tokenized keyword matching and OpenAI embedding search concurrently, returning results grouped by source file.
• Two-Stage Vector Retrieval — Source-level mean vectors shortlist candidate files, then candidate chunks are scored with parallel cosine and filtered by a similarity floor.
• One-Way Write Boundary — The API exposes queries only; the sole write path is watcher → parser → SQLite, and changed content invalidates stale embeddings automatically.
• Native MCP Tools — kura mcp serves list_rag and search_rag over stdio, and kura remote enable exposes the same tools over HTTP at /mcp.

Architecture

Full Architecture

graph TB
    Inbox[~/Kura_name folder] --> Watcher[File Watcher]
    Watcher --> SQLite[(Per-DB SQLite)]
    SQLite --> Embedder[Embedding Scheduler]
    Embedder --> OpenAI[OpenAI Embedding]
    Embedder --> Vector[Vector Cache]
    HTTP[HTTP API] --> Search[Shared Search Core]
    MCP[MCP stdio / HTTP] --> Search
    Search --> Vector
    Search --> SQLite

License

This project is licensed under the MIT LICENSE.

Author

Just open an issue to share an idea.

©️ 2026 邱敬幃 Pardn Chiu