dlightrag 1.1.6.7

DlightRAG - Dual-mode multi-modal RAG service based on LightRAG

DlightRAG

Dual-mode multimodal RAG built on LightRAG — knowledge graph + vector retrieval as a modern and unified production-ready service.

Features

• Dual multimodal RAG modes — Caption mode (parse → caption → embed) for pipeline based multimodal paradigm; Unified mode (render → multimodal embed) for modern multimodal paradigm
• Knowledge graph + vector retrieval — Fusional search based on LightRAG's foundation
• Multimodal ingestion — PDF, Word, Excel, PowerPoint, images, etc.
• Reranking — Generic LLM-based list/point-wise; Specialized rerankers support from Cohere, Jina, Aliyun, Azure Cohere; Support any additional backend via custom endpoint
• Cross-workspace federation — Query across embedding-compatible workspaces with round-robin merging
• Conversational query rewriting — Web UI rewrites follow-up messages into standalone queries via LLM
• Citation and highlighter — Answer / Retrieved contexts with source, page, citation, highlighting attribution.
• Flexible sourcing — Local filesystem, Azure Blob Storage, Snowflake
• Four interfaces — Web UI, REST API, MCP server, and Python SDK

Architecture

_{Source: docs/architecture.drawio}

Quick Start with Four Interfaces

Web UI

Click the image to watch demo (YouTube)

If you already have the REST API running (via Docker or dlightrag-api), the Web UI is available at:

http://localhost:8100/web/

Without Docker:

uv add dlightrag        # or: pip install dlightrag
cp .env.example .env    # edit .env — at minimum set DLIGHTRAG_OPENAI_API_KEY
dlightrag-api --env-file .env

Docker (Self-Hosted)

git clone https://github.com/hanlianlu/dlightrag.git && cd dlightrag
cp .env.example .env    # edit .env — at minimum set DLIGHTRAG_OPENAI_API_KEY
docker compose up

Includes PostgreSQL (pgvector + AGE), REST API (:8100), and MCP server (:8101).

Local models (Ollama, Xinference, etc.): use host.docker.internal instead of localhost in base URL settings.

curl http://localhost:8100/health

curl -X POST http://localhost:8100/ingest \
  -H "Content-Type: application/json" \
  -d '{"source_type": "local", "path": "/app/dlightrag_storage/sources"}'

curl -X POST http://localhost:8100/retrieve \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the key findings?"}'

curl -X POST http://localhost:8100/answer \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the key findings?", "stream": true}'

MCP Server (for AI Agents)

uv tool install dlightrag   # or: pip install dlightrag
cp .env.example .env        # edit .env — at minimum set DLIGHTRAG_OPENAI_API_KEY
dlightrag-mcp --env-file .env

{
  "mcpServers": {
    "dlightrag": {
      "command": "uvx",
      "args": ["dlightrag-mcp", "--env-file", "/absolute/path/to/.env"]
    }
  }
}

Tools: retrieve, answer, ingest, list_files, delete_files, list_workspaces — all with workspace isolation.

Python SDK

uv add dlightrag        # or: pip install dlightrag
cp .env.example .env    # edit .env — at minimum set DLIGHTRAG_OPENAI_API_KEY

import asyncio
from dotenv import load_dotenv
from dlightrag import RAGServiceManager, DlightragConfig

load_dotenv()  # load .env

async def main():
    config = DlightragConfig()
    manager = RAGServiceManager(config)

    await manager.aingest(workspace="default", source_type="local", path="./docs")

    result = await manager.aretrieve(query="What are the key findings?")
    print(result.contexts)

    result = await manager.aanswer(query="What are the key findings?")
    print(result.answer)

asyncio.run(main())

Requires PostgreSQL with pgvector + AGE, or JSON fallback for development (see Configuration).

API Reference

Request and response structures for ingest, retrieve, and answer across all interfaces. See docs/response-schema.md for the full reference — ingestion parameters, retrieval contexts, sources, media, SSE streaming, citations, and multimodal queries.

Configuration

All settings via DLIGHTRAG_ env vars, .env file, or constructor args. See .env.example for the full reference.

Priority: constructor args > env vars > .env file > defaults

RAG Mode

The first decision — determines your ingestion pipeline, model requirements, and retrieval behavior.

Mode	Pipeline	Best for
caption (default)	Document parsing → VLM captioning → text embedding → KG	Text-heavy documents, structured elements
unified	Page rendering → multimodal embedding → VLM entity extraction → KG	Visually rich documents (charts, diagrams, complex layouts)

Caption mode parsers (DLIGHTRAG_PARSER):

Parser	Description
mineru (default)	MinerU PDF parser — fast, good for text-heavy documents
docling	Docling parser — alternative structure-aware parser
vlm	VLM-based OCR — renders pages and uses vision model to extract structured content; no external parser dependency, requires VISION_MODEL

All caption mode parsers use Docling's HybridChunker for structure-aware chunking.

Model usage by stage:

Stage	Caption	Unified
Image captioning	VISION_MODEL ¹	VISION_MODEL
Table / equation captioning	CHAT_MODEL	—
Entity extraction	CHAT_MODEL	CHAT_MODEL
Embedding	EMBEDDING_MODEL	EMBEDDING_MODEL (multimodal)
Rerank	RERANK_* via LightRAG	VISION_MODEL ² or RERANK_* API
Answer generation	CHAT_MODEL via AnswerEngine	VISION_MODEL via AnswerEngine (sees page images)

¹ Falls back to CHAT_MODEL if vision model not configured. ² When RERANK_BACKEND=llm (pointwise VLM scoring).

For unified mode, set DLIGHTRAG_RAG_MODE=unified and point embedding/vision at multimodal models:

DLIGHTRAG_RAG_MODE=unified
DLIGHTRAG_EMBEDDING_MODEL=Qwen3-VL-Embedding    # must be multimodal
DLIGHTRAG_EMBEDDING_DIM=4096
DLIGHTRAG_VISION_MODEL=qwen3-vl-32b

Limitations: Snowflake is text-only (no visual embedding). A workspace is locked to one mode after first ingestion. Page images ~3-7 MB/page at 250 DPI.

Providers

Variable	Default	Description
DLIGHTRAG_LLM_PROVIDER	openai	openai, azure_openai, anthropic, google_gemini, qwen, minimax, xinference, openrouter, ollama, voyage
DLIGHTRAG_EMBEDDING_PROVIDER	(follows llm_provider)	Override embedding provider
DLIGHTRAG_VISION_PROVIDER	(follows llm_provider)	Override vision provider
DLIGHTRAG_EMBEDDING_MODEL	text-embedding-3-large	Embedding model

Each provider uses its own API key. For Ollama, use openai provider with DLIGHTRAG_OPENAI_BASE_URL pointing to Ollama.

Storage Backends

Variable	Default	Options
DLIGHTRAG_VECTOR_STORAGE	PGVectorStorage	PGVectorStorage, MilvusVectorDBStorage, NanoVectorDBStorage, ...
DLIGHTRAG_GRAPH_STORAGE	PGGraphStorage	PGGraphStorage, Neo4JStorage, NetworkXStorage, ...
DLIGHTRAG_KV_STORAGE	PGKVStorage	PGKVStorage, JsonKVStorage, RedisKVStorage, ...
DLIGHTRAG_DOC_STATUS_STORAGE	PGDocStatusStorage	PGDocStatusStorage, JsonDocStatusStorage, ...

Note: When using PostgreSQL backends, LightRAG maps its internal namespace names to different table names (e.g. text_chunks → LIGHTRAG_DOC_CHUNKS, full_docs → LIGHTRAG_DOC_FULL). DlightRAG's unified mode adds a visual_chunks table via its own KV storage.

Workspaces

Each workspace has its own knowledge graph, vector store, and document index. DLIGHTRAG_WORKSPACE (default: default) is automatically bridged to backend-specific env vars — no manual setup needed.

Backend type	Isolation mechanism
PostgreSQL (PG*)	workspace column / graph name in same database
Neo4j / Memgraph	Label prefix
Milvus / Qdrant	Collection prefix
MongoDB / Redis	Collection scope
JSON / Nano / NetworkX / Faiss	Subdirectory under working_dir/<workspace>/

Reranking

Variable	Default	Description
DLIGHTRAG_RERANK_BACKEND	llm	llm, cohere, jina, aliyun, azure_cohere
DLIGHTRAG_RERANK_MODEL	(backend default)	Model name sent to the endpoint
DLIGHTRAG_RERANK_BASE_URL	(provider default)	Custom endpoint URL for any compatible service
DLIGHTRAG_RERANK_API_KEY	—	API key (falls back to provider-specific keys)

Backend	Default model	Key
llm	(follows CHAT_MODEL)	(follows LLM_PROVIDER)
cohere	rerank-v4.0-pro	DLIGHTRAG_COHERE_API_KEY
jina	jina-reranker-v3	DLIGHTRAG_JINA_API_KEY
aliyun	qwen3-rerank	DLIGHTRAG_ALIYUN_RERANK_API_KEY
azure_cohere	Cohere-rerank-v4.0-pro	DLIGHTRAG_AZURE_COHERE_API_KEY

Point any backend at a local reranker (Xinference, LiteLLM, etc.) via RERANK_BASE_URL + RERANK_MODEL.

REST API

Method	Endpoint	Description
POST	/ingest	Ingest from local, Azure Blob, or Snowflake
POST	/retrieve	Contexts + sources (no LLM answer)
POST	/answer	LLM answer + contexts + sources (stream: true for SSE)
GET	/files	List ingested documents
DELETE	/files	Delete documents
GET	/api/files/{path}	Serve/download a file (local: stream, Azure: 302 SAS redirect)
GET	/workspaces	List available workspaces
GET	/health	Health check with storage status

All write endpoints accept optional workspace; read endpoints accept workspaces list for cross-workspace federated search. Set DLIGHTRAG_API_AUTH_TOKEN to enable bearer auth.

Development

git clone https://github.com/hanlianlu/dlightrag.git && cd dlightrag
cp .env.example .env && uv sync
docker compose up -d                # PostgreSQL + API + MCP
docker compose up postgres -d       # PostgreSQL only

uv run pytest tests/unit            # unit tests (no external services)
uv run pytest tests/integration     # integration tests (requires PostgreSQL)
uv run ruff check src/ tests/ scripts/ --fix && uv run ruff format src/ tests/ scripts/

License

Apache License 2.0 — see LICENSE.

---

Built by HanlianLyu. Contributions welcome!