Fusion Query Engine — Dify Plugin

LlamaIndex-based tool plugin for Dify that performs windowed COMPACT synthesis
over pre-fetched context (Qdrant points or plain text), with optional chat-history
condensing.

Author: gitpoldan. Provider id: .

Features

• Two modes: (enrich query) and (generate answer)
• Pre-fetched context — no Qdrant calls inside the plugin; nodes from upstream workflow
• Chat history — optional JSON for multi-turn condensing
• Two LLM selectors — separate models for expansion and answer steps

Setup

1. Install the plugin in Dify (Marketplace upload or remote debug).
2. Add the Fusion Query Engine tool to a Workflow or Chatflow.
3. Configure model_expansion and model_answer (LLM model selectors from your workspace).
4. Wire inputs:
   • query — user question ( or LLM output)
   • source_nodes — array from Knowledge Retrieval or a Code node (Qdrant-style points)
   • context_text — plain-text fallback when is empty
   • chat_history — optional JSON string of prior turns (see Usage)

No external API credentials beyond LLMs already configured in Dify.

Usage

Two-step RAG pipeline:

1. Run tool with → enriched query in
2. Run Knowledge Retrieval with enriched query
3. Run tool with → final answer in

Outputs: , , (expansion mode only).

Example element:

Requires Dify >= 1.14.0.

Privacy

Queries and context are sent to LLM providers configured in your workspace. See PRIVACY.md.

Support

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

Russian documentation: readme/README_ru_RU.md.