:) OBSIZEN

Transform your Obsidian vault into an intelligent AI knowledge companion.

OBSIZEN is an open-source Retrieval-Augmented Generation (RAG) system designed specifically for Obsidian users.

It combines semantic search, keyword search, metadata-aware retrieval, and reranking to provide grounded answers from your notes.

---

:D Features

Features

• Hybrid Retrieval (Vector + BM25)
• Metadata-Aware Retrieval
• Filename Retrieval
• Source Deduplication
• CrossEncoder Reranking
• Incremental Indexing
• Cached Models for Fast Startup
• Config-Driven Architecture
• OpenRouter Integration

---

Architecture

Obsidian Vault
      │
      ▼
Markdown Loader
      │
      ▼
Markdown Header Chunking
      │
      ▼
Embedding Generation
      │
      ▼
ChromaDB
      │
      ▼
Hybrid Retrieval
(Vector + BM25 + Filename Search)
      │
      ▼
Metadata Boosting
      │
      ▼
Source Deduplication
      │
      ▼
CrossEncoder Reranker
      │
      ▼
Context Builder
      │
      ▼
DeepSeek via OpenRouter
      │
      ▼
Grounded Answer

Usage

Build index:

python src/main.py

Ask questions:

Ask Away!

What are my YC ideas? What did I learn about FastAPI? Summarize my ML notes.

Future Roadmap: Logging Tests Error Handling

FastAPI Backend

React Web UI

Local LLM Support

Tech Stack

LangChain
ChromaDB
SentenceTransformers
CrossEncoder
OpenRouter
Python

Contribution

Pull requests are welcome.

License

MIT License