Skip to main content

RAG Pipeline

Build a complete RAG pipeline: load documents, embed them in ZeroDB, search by meaning, and generate answers.

Prerequisites​

pip install langchain-zerodb langchain-community

Full Pipeline​

from langchain_zerodb import ZeroDBVectorStore

# 1. Connect to ZeroDB
store = ZeroDBVectorStore(
    api_key="your-api-key",
    project_id="your-project-id",
)

# 2. Load and embed documents (embeddings are free)
docs = [
    "ZeroDB is a persistent knowledge layer for AI agents.",
    "It provides vector search, agent memory, and file storage.",
    "GraphRAG combines vector search with knowledge graph traversal.",
    "ZeroMemory supports working, episodic, and semantic memory tiers.",
    "MCP servers let agents access ZeroDB tools directly.",
]
store.add_texts(docs)

# 3. Search by meaning
results = store.similarity_search("How does memory work?", k=3)
for doc in results:
    print(f"  - {doc.page_content}")

# 4. Use as a retriever in a QA chain
from langchain.chains import RetrievalQA
from langchain_community.llms import Ollama

qa = RetrievalQA.from_chain_type(
    llm=Ollama(model="llama3"),
    retriever=store.as_retriever(search_kwargs={"k": 3}),
)

answer = qa.run("What memory tiers does ZeroMemory support?")
print(f"Answer: {answer}")

Expected Output​

  - ZeroMemory supports working, episodic, and semantic memory tiers.
  - ZeroDB is a persistent knowledge layer for AI agents.
  - It provides vector search, agent memory, and file storage.
Answer: ZeroMemory supports three memory tiers: working, episodic, and semantic.

What to Try Next​