ragstack-ai-knowledge-store

RAGStack Graph Store

Hybrid Graph Store combining vector similarity and edges between chunks.

Usage

• Pre-process your documents to populate metadata information.
• Create a Hybrid GraphStore and add your LangChain Documents.
• Retrieve documents from the GraphStore.

Populate Metadata

The Graph Store makes use of the following metadata fields on each Document:

• content_id: If assigned, this specifies the unique ID of the Document. If not assigned, one will be generated. This should be set if you may re-ingest the same document so that it is overwritten rather than being duplicated.
• links: A set of Links indicating how this node should be linked to other nodes.

Hyperlinks

To connect nodes based on hyperlinks, you can use the HtmlLinkExtractor as shown below:

from ragstack_knowledge_store.langchain.extractors import HtmlLinkExtractor

html_link_extractor = HtmlLinkExtractor()

for doc in documents:
    doc.metadata["content_id"] = doc.metadata["source"]

    # Add link tags from the page_content to the metadata.
    # Should be passed the HTML content as a string or BeautifulSoup.
    add_links(doc,
        html_link_extractor.extract_one(HtmlInput(doc.page_content, doc.metadata["source_url"])))

Store

import cassio
from langchain_openai import OpenAIEmbeddings
from ragstack_knowledge_store import GraphStore

cassio.init(auto=True)

graph_store = GraphStore(embeddings=OpenAIEmbeddings())

# Store the documents
graph_store.add_documents(documents)

Retrieve

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o")

# Retrieve and generate using the relevant snippets of the blog.
from langchain_core.runnables import RunnablePassthrough
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate

# Depth 0 - don't traverse edges. equivalent to vector-only.
# Depth 1 - vector search plus 1 level of edges
retriever = graph_store.as_retriever(k=4, depth=1)

template = """You are a helpful technical support bot. You should provide complete answers explaining the options the user has available to address their problem. Answer the question based only on the following context:
{context}

Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)

def format_docs(docs):
    formatted = "\n\n".join(f"From {doc.metadata['content_id']}: {doc.page_content}" for doc in docs)
    return formatted


rag_chain = (
    {"context": retriever | format_docs, "question": RunnablePassthrough()}
    | prompt
    | llm
    | StrOutputParser()
)

Development

poetry install --with=dev

# Run Tests
poetry run pytest