Qdrant Hybrid Retriever (gitpoldan)

Qdrant hybrid retrieval plugin for Dify: dense + sparse (BM25) with DBSF/RRF
fusion, optional MMR, tenant automode/ratios, payload filters, and optional reranking.

Author: gitpoldan. Provider id: .

Features

Hybrid search — dense vectors + BM25 sparse via official
Fusion — DBSF (default) or RRF; optional MMR on dense branch
Tenant modes — automode median prune and/or weighted quota merge
Filters — , , , JSON
Rerank — optional Dify rerank model or local fallback
LLM-friendly outputs — , , ,

Setup

Deploy Qdrant 1.15+ (MMR requires 1.15+).
Install the plugin and open Tool provider → Qdrant Hybrid Retriever.
Configure credentials:

Field	Required	Description
	Yes	Qdrant URL, e.g.
	No	Required for Qdrant Cloud
	No	Optional JSON HTTP headers

In the tool node, set collection, text (query), and embedding_model_config
(embedding model from your Dify workspace).

Usage

Add Hybrid Retriever to a Workflow node. Key parameters:

, , ( / )
, , , (JSON)
,
— expand to all points from top-K sources (optional)

Outputs: use or as LLM context downstream.

Privacy

Connects to your Qdrant cluster; embedding/rerank via Dify models. See PRIVACY.md.

Support

Source: https://github.com/gitpoldan/dify/tree/main/polden-plugins/qdrant-hybrid
GitHub Issues: https://github.com/gitpoldan/dify/issues
Email: [email protected]

Russian documentation: readme/README_ru_RU.md.