langchain-machinelibrary

Machine Library for LangChain: a retriever and agent tools for citable search across scholarly papers, books, patents, standards and Wikipedia (plus Reddit, Telegram and Discord). Results carry a canonical source_uri, so answers can cite sources a reader can open and check.

Install

pip install langchain-machinelibrary
export MACHINELIBRARY_API_KEY=ml_...   # https://machinelibrary.ai/keys

Retriever (RAG)

from langchain_machinelibrary import MachineLibraryRetriever

retriever = MachineLibraryRetriever(k=5)
for doc in retriever.invoke("perovskite solar cell stability under humidity"):
    print(doc.metadata["title"], doc.metadata["source"])
    print(doc.page_content)

Agent tools

from langchain.agents import create_agent
from langchain_machinelibrary import machinelibrary_tools

agent = create_agent(
    "anthropic:claude-sonnet-5-5",
    tools=machinelibrary_tools(),
    system_prompt="Answer from Machine Library sources and cite each claim's source_uri.",
)
agent.invoke({"messages": [{"role": "user", "content": "What limits lithium-metal anodes?"}]})

Billing and limits

Each search or read is charged to the key's account; see pricing. Retrieved text is evidence to review, not proof that a generated answer is correct. Coverage is broad but not exhaustive.