BM25 RAG

In this example, we will show how to use the Elasticsearch as in DB-GPT RAG Storage. Using a Elasticsearch database to implement RAG can, to some extent, alleviate the uncertainty and interpretability issues brought about by Elasticsearch database retrieval.

Install Dependencies

First, you need to install the dbgpt elasticsearch storage library.

uv sync --all-packages --frozen \
--extra "base" \
--extra "proxy_openai" \
--extra "rag" \
--extra "storage_elasticsearch" \
--extra "dbgpts"

Prepare Elasticsearch

Prepare Elasticsearch database service, reference-Elasticsearch Installation .

Elasticsearch Configuration

Set rag storage variables below in configs/dbgpt-bm25-rag.toml file, let DB-GPT know how to connect to Elasticsearch.

[rag.storage]
[rag.storage.full_text]
type = "ElasticSearch"
uri = "127.0.0.1"
port = "9200"

Then run the following command to start the webserver:

uv run python packages/dbgpt-app/src/dbgpt_app/dbgpt_server.py --config configs/dbgpt-bm25-rag.toml

Optionally

uv run python packages/dbgpt-app/src/dbgpt_app/dbgpt_server.py --config configs/dbgpt-bm25-rag.toml