langchain-ckg

LangChain retriever backed by Compact Knowledge Graphs (CKG) — structured, deterministic, SHA-256 anchored domain knowledge over MCP.

Why

RAG retrieves document chunks. CKG retrieves typed dependency graphs. The difference:

Install

pip install langchain-ckg

Usage

Local retriever (free, bundled graphs)

from langchain_ckg import CKGRetriever
from langchain_openai import ChatOpenAI
from langchain.chains import RetrievalQA

retriever = CKGRetriever(domain="nvidia-nemoclaw", depth=3)
qa = RetrievalQA.from_chain_type(llm=ChatOpenAI(), retriever=retriever)
result = qa.invoke("How does NemoClaw handle CUDA kernel fusion?")

Hosted retriever (rate-gated, upgradeable)

from langchain_ckg import CKGHostedRetriever

# Free: 50 calls/day
retriever = CKGHostedRetriever(domain="nvidia-nemoclaw")

# Unlimited: $10/yr at graphifymd.com/pricing
retriever = CKGHostedRetriever(domain="nvidia-nemoclaw", license_key="your-key")

Available domains

97 domains including NVIDIA AI, NemoClaw, Salesforce AgentForce, Nemotron, and more.

uvx ckg-mcp  # then call list_domains

Links

• graphifymd.com
• Benchmark dataset
• PyPI: ckg-mcp
• Pricing