Serpent

A modular RAG framework for C++

---

Serpent is a small C++ framework for building RAG (retrieval-augmented generation) pipelines and AI-powered applications.

Features

• Modular Design: Swap components via clean interfaces
• Hybrid Search: hnswlib (dense) + BM25 (sparse) + RRF fusion
• Dual LLM Support: Local models (llama.cpp) + Cloud APIs (OpenRouter)
• Document Loaders: PDF, DOCX, and plain text

---

Architecture

graph TB
    subgraph Loaders["Document Loaders"]
        TXT["TextLoader"]
        PDF["PdfLoader"]
        DOCX["DocxLoader"]
    end

    subgraph Chunking["Text Chunking"]
        RC["RecursiveChunker"]
    end

    subgraph Core["Core"]
        DS["MemoryDocumentStore"]
        SQL["SqliteStore"]
    end

    subgraph Retrieval["Retrieval"]
        HNSW["HNSWStore<br/>(Dense Vectors)"]
        BM25["BM25Index<br/>(Sparse)"]
        HR["HybridRetriever"]
        RRF["RRF Fusion"]
    end

    subgraph Providers["LLM Providers"]
        LLAMA["LlamaCppProvider<br/>(Local)"]
        OR["OpenRouterProvider<br/>(Cloud)"]
    end

    subgraph Utils["Utilities"]
        MM["ModelManager"]
    end

    TXT & PDF & DOCX --> RC
    RC --> DS
    DS --> SQL
    DS --> HNSW & BM25
    HNSW & BM25 --> HR
    HR --> RRF
    RRF --> LLAMA & OR
    MM --> LLAMA

Loading

---

Quickstart

To be added.

---

Documentation

To be added.

---

Contributing

See CONTRIBUTING.md for details on how to make a contribution.

License

Serpent is available under either of the following licenses:

• MIT License (LICENSE-MIT)
• Apache License, Version 2.0 (LICENSE-APACHE)

Acknowledgements

• The logo is from here with some modifications.