lfx-nextplaid

NextPlaid multi-vector vector store (plus its companion vLLM multivector embeddings) as a standalone Langflow Extension Bundle.

NextPlaid stores each document as a matrix of token embeddings rather than a single vector, enabling ColBERT/ColPali-style late interaction (MaxSim scoring) for significantly higher retrieval quality on semantic search. The bundle ships two components:

NextPlaidVectorStoreComponent — vector store backed by a running NextPlaid server via the langchain-plaid client. Supports text (ColBERT) and image (ColPali) ingestion with full upsert semantics via stable document IDs.
VllmMultivectorEmbeddingsComponent — produces the token-matrix embeddings NextPlaid ingests by calling vLLM's /pooling endpoint with task: token_embed. Compatible with ColBERT models such as answerdotai/answerai-colbert-small-v1.

Requirements

A running NextPlaid server.
A running vLLM server with a ColBERT-compatible model loaded via --runner pooling.

Install

bash

pip install lfx-nextplaid

The bundle is registered automatically via the langflow.extensions entry-point. After install, restart your Langflow server; the components appear in the palette under the nextplaid bundle group.

Develop

bash

cd src/bundles/nextplaid
pip install -e .
lfx extension validate src/lfx_nextplaid

Manifest

The extension manifest is shipped at src/lfx_nextplaid/extension.json and points at the bundle at components/nextplaid. Components register under the canonical namespaced IDs ext:nextplaid:NextPlaidVectorStoreComponent@official and ext:nextplaid:VllmMultivectorEmbeddingsComponent@official.

Migration

Saved flows referencing the legacy class names or the old import paths under lfx.components.nextplaid.* / lfx.components.vllm.VllmMultivectorEmbeddingsComponent are rewritten to the new namespaced IDs by the migration table in src/lfx/src/lfx/extension/migration/migration_table.json.