graphrag-light

Welcome to graphrag-light. An easy to start, GCP native, graphrag implementation.

Installation Steps

1. Run in terminal to create a python venv:

python3 -m venv .venv-graphrag-lite 

2. graphrag-lite uses graph2nosql, a simple python interface to store and manage knowledge graphs in your NoSQL DB of choice (no graphdb needed, you heard correctly!).

Add graph2nosql into your project folder for local installation. No pip distribution available (yet).

git clone https://github.com/jakobap/graph2nosql.git

3. Install graphrag-lite dependencies

pip install -r ./graphrag_lite/requirements.txt

4. This repo implements graphrag as paralelized architecture on GCP. Large parts of the indexing and query steps are processed by two stateless microservices. These are defined in stateless-comm-reporter and stateless-context-processor. Both directories contain Makefiles to automate the build and deployment.

To deploy:

• Generate one or multiple service account key and place it/them in stateless-comm-reporter and stateless-context-processor.
• Setup an .env file with the following environment variables and place it in stateless-comm-reporter and stateless-context-processor:

GCP_CREDENTIAL_FILE=""
GCP_PROJECT_ID=""
GCP_PROJECT_NUMBER =""
GCP_REGION =""
GCP_MULTIREGION = ""
DOCUMENT_AI_PROCESSOR_ID =""
DOCUMENT_AI_PROCESSOR_VERSION = ""
LANGFUSE_SECRET_KEY = 
LANGFUSE_PUBLIC_KEY = 
LANGFUSE_HOST=
FIRESTORE_DB_ID=""
NODE_COLL_ID=""
COMM_COLL_ID=""
EDGES_COLL_ID=""
SCHEDULER_PUBSUB_ID=""
COMMUNITY_WL_PUBSUB=""
QUERY_FS_DB_ID=""
QUERY_FS_INT__RESPONSE_COLL=""
RAW_PDFS_BUCKET_NAME=""

• Run

make all