🦜️🔗 LangChain NextPlaid

This repository contains 1 package with NextPlaid integrations with LangChain:

• langchain-plaid (WIP)

Overview

NextPlaid is a high-performance ColBERT-style multi-vector search engine built in Rust. This package provides a LangChain VectorStore integration, enabling late-interaction retrieval with full metadata filtering support.

Installation

pip install langchain-plaid

Usage

from langchain_plaid import NextPlaidVectorStore

vectorstore = NextPlaidVectorStore(
    url="http://localhost:8080",
    index_name="my_index",
    embedding=your_embedding_model,
)

# Add documents
vectorstore.add_documents(documents)

# Similarity search
results = vectorstore.similarity_search("query", k=4)

# Similarity search with metadata filter
results = vectorstore.similarity_search(
    "query",
    k=4,
    filter={"category": "science"},
)

Running the NextPlaid Server

cargo run --release -p next-plaid-api -- --index-dir /tmp/indices --port 8080

See the next-plaid repository for full server documentation.

Development

# Install dependencies
cd libs/plaid
pip install -e ".[dev]"

# Run integration tests (requires a running NextPlaid server)
python -m pytest tests/integration_tests/