memvid-rs 2.1.0

High-performance AI memory library with full-text and semantic search (Rust bindings for memvid v2)

memvid-rs (Python Bindings)

Memvid-rs provides high-performance Python bindings for memvid-core v2, the engine behind the memvid ecosystem. It is a crash-safe, single-file AI memory designed for RAG and long-term agent memory.

Attribution: This project wraps memvid-core with PyO3 bindings. API is designed to be compatible with memvid-sdk.

🚀 Key Features

• Single-File Memory: All data (text, index, metadata) stored in portable .mv2 format.
• Full-Text & RAG: BM25 ranking via embedded Tantivy index with built-in RAG (ask) support.
• Extreme Performance: Rust-core ensures sub-millisecond lookups and efficient batch ingestion.
• Crash-Safe: WAL-based append-only architecture protects your data.
• Compatibility: Drop-in replacement for memvid-sdk Python users needing more speed.

📦 Installation

pip install memvid-rs

💻 Quick Start

from memvid_rs import use

# Use 'auto' mode to create or open memory
with use("ai_agent", "data.mv2", mode="auto") as mv:
    # Add document with metadata and auto-indexing
    mv.put(
        text="Memvid-RS is a ultra-fast Rust implementation.",
        title="Performance Note",
        label="tech",
        tags=["rust", "fast"]
    )
    
    # Batch ingestion (100x faster than individual put)
    mv.put_many([
        {"text": "Sample doc 1", "title": "Doc 1"},
        {"text": "Sample doc 2", "title": "Doc 2"},
    ])
    
    mv.commit()
    
    # Full-text search
    results = mv.find("rust implementation", k=5)
    print(f"Search results: {results['hits']}")
    
    # RAG Question Answering
    answer = mv.ask("What is Memvid-RS?")
    print(f"AI Answer: {answer['answer']}")

🔧 API Reference

Top-level Functions

• use(profile, path, mode="auto", read_only=False): Flexible entry point.
• create(path, capacity_gb=1): Create new memory.
• open(path): Open existing memory.

Memory / MemvidMemory Methods

Method	Description
put(text, title, ...)	Insert document with metadata
put_many([docs])	Batch insert multiple documents
find(query, k)	Full-text search (BM25)
ask(question, k)	RAG-based question answering
append(text)	Low-level text append
commit() / seal()	Flush writes to disk
stats()	Get frame count and size in bytes
timeline(limit)	Retrieve recent documents

⚠️ Breaking Changes from v1

• QR video encoding removed - v2 uses binary .mv2 format for 1000x better density.
• API redesigned - Now follows the memvid-sdk v2 patterns.

📄 License

MIT License