getmnemo-langchain

LangChain adapter for Mnemo Memory — drop-in conversational memory and retriever backed by the getmnemo SDK.

Install

pip install getmnemo-langchain

Requires getmnemo>=0.2.

Quickstart

from getmnemo import Mnemo
from langchain.chains import ConversationChain
from langchain_openai import ChatOpenAI

from langchain_getmnemo import MnemoMemory

mem_client = Mnemo(api_key="prfly_live_...", default_container_tag="user:jane")
chain = ConversationChain(
    llm=ChatOpenAI(model="gpt-4o-mini"),
    memory=MnemoMemory(client=mem_client, top_k=5),
)

print(chain.invoke({"input": "Remember that my favourite framework is FastAPI."}))
print(chain.invoke({"input": "What is my favourite framework?"}))

Every call is scoped to a container — a "<type>:<id>" string such as "user:jane". Set it once on the client (default_container_tag=) or per component (container_tag=):

memory = MnemoMemory(client=mem_client, container_tag="user:jane")

Retriever for RAG

from langchain_getmnemo import MnemoRetriever

retriever = MnemoRetriever(client=mem_client, top_k=8)
docs = retriever.invoke("show me what I said about deployments")
for doc in docs:
    print(doc.metadata.get("score"), doc.page_content)

What you get

• MnemoMemory — implements langchain_core.memory.BaseMemory. Auto-saves every user / AI turn and rehydrates the top-K relevant prior turns by semantic search. An optional namespace partitions threads that share one container.
• MnemoRetriever — implements langchain_core.retrievers.BaseRetriever. Returns LangChain Document objects with the memory's metadata plus score, memory_id, and (for document-derived hits) document_id.

License

MIT — see LICENSE.