mnemosyne-langchain

LangChain and LangGraph integration for the Mnemosyne memory substrate. MnemosyneChatHistory, MnemosyneRetriever, and MnemosyneMemory classes, with hybrid BM25+vector recall, per-session thread tracking, automatic redaction, and TTL support.

Install

pip install mnemosyne-langchain
# or with the type stubs for full IDE support:
pip install "mnemosyne-langchain[langchain]"

Quick start

import os
from mnemosyne_langchain import MnemosyneChatHistory, MnemosyneRetriever, MnemosyneMemory

os.environ["MNEMOSYNE_API_KEY"] = "mn_..."

# Standalone chat history
history = MnemosyneChatHistory(session_id="alice-1", container_tag="default")
history.add_user_message("Hello!")
history.add_ai_message("Hi, how can I help?")
for m in history.messages:
    print(m)

# Retriever
retriever = MnemosyneRetriever(container_tag="default")
docs = retriever.invoke("Python best practices")
for d in docs:
    print(d.page_content, d.metadata)

# Full memory object
memory = MnemosyneMemory(session_id="alice-1")
vars = memory.load_memory_variables({"input": "what did I ask before?"})
print(vars)
memory.save_context({"input": "thanks!"}, {"output": "you're welcome"})
memory.clear()

With a LangChain chain

from langchain_openai import ChatOpenAI
from langchain.chains import LLMChain
from langchain.prompts import PromptTemplate
from mnemosyne_langchain import MnemosyneMemory

llm = ChatOpenAI(model="gpt-4")
prompt = PromptTemplate(
    input_variables=["history", "context", "input"],
    template="Context:\n{context}\n\nHistory:\n{history}\n\nUser: {input}\nAssistant:",
)
memory = MnemosyneMemory(session_id="alice-1", k=3)
chain = LLMChain(llm=llm, prompt=prompt, memory=memory)

chain.invoke({"input": "remind me what we discussed about caching"})

License

MIT © Geass Labs