tpt-raglite

The SQLite for AI — Blazing-fast, embeddable, zero-configuration RAG engine in pure Rust.

Overview

tpt-raglite replaces the fragmented vector-DB + embedding-server + chunking-framework stack with a single, statically compiled library. It handles document parsing, local ONNX-based embeddings, smart chunking, and disk-backed vector indexing — entirely offline, with zero external infrastructure.

Features

• Zero-config & offline-first — No Docker, no API keys, no cloud dependencies
• Embeddable, not a server — Runs in the same process as your application
• Multi-language — Rust core with C ABI for Python (pip install), JavaScript (npm install), and C++
• Memory-efficient — Memory-mapped HNSW index handles millions of vectors
• Fast — Query latency < 10ms at 1M vectors on modern CPUs

Quick Start

Rust

use tpt_rag_core::RAGLite;
use std::path::Path;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let mut rag = RAGLite::open(Path::new("./my_db"))?;

    // Add a document
    rag.add_document(Path::new("manual.pdf"), &["docs"])?;

    // Add text directly
    rag.add_text("Some important information.", &Default::default())?;

    // Query
    let results = rag.query("How do I configure the engine?", 5)?;
    for r in &results {
        println!("[{:.2}] {}", r.score, r.text);
    }
    Ok(())
}

Python

from tpt_raglite import RAGLite

rag = RAGLite("./my_db")
rag.add_files(["manual.pdf", "notes.md"], tags=["docs"])
results = rag.query("How do I configure the engine?")
for r in results:
    print(f"[{r.score:.2f}] {r.text}")

C

#include "tpt_raglite.h"

int main() {
    tpt_rag_handle* rag = tpt_rag_create("./my_c_db");
    tpt_rag_add_file(rag, "manual.pdf");

    tpt_rag_result* results = NULL;
    int count = tpt_rag_query(rag, "How to configure?", 5, &results);

    // Process results...

    tpt_rag_free_results(results, count);
    tpt_rag_destroy(rag);
    return 0;
}

Installation

From source (Rust)

cargo build --release

Python

pip install tpt-raglite

JavaScript / WASM

npm install tpt-raglite-wasm

Supported Formats

Format	Extension
Plain text	.txt
Markdown	.md, .markdown, .mdown
HTML	.html, .htm
PDF	.pdf

Architecture

tpt-rag-core/     # Pure Rust: parsing, chunking, ONNX embedding, HNSW index, SQLite metadata
tpt-rag-ffi/      # C ABI boundary with opaque handles
tpt-rag-py/       # PyO3 Python wrapper
tpt-rag-wasm/     # wasm-bindgen JS/Browser wrapper

Requirements

• Models: Place a quantized all-MiniLM-L6-v2.onnx model in the models/ directory or alongside your database path.
• The model provides 384-dimensional embeddings with < 100MB RAM usage.

License

MIT OR Apache-2.0 — (c) TPT Solutions