Practical overview
When it fits
Use Memgraphrag when you want to prototype a self-hosted GraphRAG pipeline that connects source text with extracted relations and abstract schemas, and you can configure the language model and embedding components it requires.
Who it is for
- Developers prototyping graph-based retrieval and question-answering pipelines.
- Researchers exploring multi-agent GraphRAG, ontology induction or memory structures connecting evidence to facts and schemas.
Documented uses
- Index a UTF-8 text corpus into linked passage, fact and schema layers.
- Construct graph artifacts from extracted knowledge and resolve connected conflicts during construction.
- Retrieve relevant material using embedding similarity and Personalized PageRank.
- Run dataset question answering and save answers, retrieved passages and run metadata.
Requirements & limits
The public GitHub repository identifies the project as MIT-licensed. The documented workflow is self-hosted and command-line based.
- The pipeline requires configuration of an OpenAI-compatible LLM endpoint and a local Hugging Face embedding model for its online setup.
- Use identical LLM and embedding model identifiers for indexing and retrieval because those values identify saved graph and embedding stores.
- Indexing takes UTF-8 plain text; question answering uses a JSON question dataset.
- Offline OpenIE is described as requiring vLLM or llama-factory; optional embedding backends may require additional packages.
- The repository documents Python 3.10 or later as recommended.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Index a text corpus and run dataset question answering
Build the documented memory graph from a corpus, then retrieve answers for a JSON question dataset.
Before you start
- Python 3.10 or later is recommended.
- A UTF-8 plain-text corpus and a JSON question dataset.
- An OpenAI-compatible LLM endpoint and API key, plus a local Hugging Face embedding model.
- Install repository dependencies; configure the LLM name, endpoint URL and embedding model. Use matching LLM and embedding model identifiers for indexing and retrieval.
Steps
Clone the repository and install its requirements in a Python environment.
Configure the API key, LLM name, endpoint URL and local embedding model as described in the README.
From the repository root, run code/index.py with the corpus path, save directory, model settings, chunk size, chunk overlap and artifact mode.
Run code/retrieval_dataset_test.py with the question dataset, saved index directory, output path and matching model settings.
Expected outcome
The indexing stage writes memory and graph artifacts; the QA stage saves a JSON result containing a run summary, configuration, solutions, raw responses, metadata and retrieved passages.
Guide limitations
- The documented corpus input is UTF-8 plain text; the QA runner expects its documented question JSON structure.
- Offline OpenIE requires additional installation of vLLM or llama-factory.
- The README provides launch templates but says to adjust interpreter, paths, endpoint, model and GPU settings before using them.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 12 October 2026
- Original website title
- XMUDeepLIT/MemGraphRAG · Official homepage title · retained in its source language