Pre-release version. Not ready for production use.
Flask-Langchain is a Flask extension that provides a simple interface for using Langchain with Flask. Currently, it provides an SQLAlchemy based memory class for storing conversation histories, on a per-user or per-session basis, and a ChromaVectorStore class for storing document vectors (per-user only).
To use, simply create a LangchainFlaskMemory object, passing in your Flask app and SQLAlchemy db object. Then, use the LangchainFlaskMemory object to create a ConversationFlaskMemory object, which can be passed to a Langchain chain as the memory parameter.
Set the user id using the set_user_id method in your app's login callback. If no user id is set, the session id will be used.
Flask-Langchain adds a session and conversation id to the Flask session object, along with a user id if provided. After the extension is initialized, the LangchainFlaskMemory object exposes chat_memory and chroma_vector_store properties which can be used to create ConversationFlaskMemory and ChromaVectorStore objects, respectively. Messages are retrieved by conversation id, and document collections are organized by user id. If no user id is provided, the session id is used instead, allowing for anonymous users (note that some LLM providers, including OpenAI, discourage the use of their APIs in unauthenticated contexts).
These classes inherit from the Langchain BaseChatMemory and ChromaVectorStore classes, and can be used in the same way.
Coming soon - for now, clone the repo and import the Flask-Langchain.extension module.
Requires Flask-SQLAlchemy.
- Incorrect formatting of conversation history (as a string representation of a list of BaseMessages).
- After it is initialized, you have to access the LangchainFlaskMemory object using
current_app.extensions['langchain_chat_memory']- this could be improved - Test coverage may be incomplete
- Supports in-memory Chroma database only
from flask import Flask, request, jsonify
from flask_sqlalchemy import SQLAlchemy
import os
from langchainmemory import LangchainFlaskMemory, ConversationFlaskMemory
from langchain.llms import OpenAI
from langchain.chains import ConversationChain
app = Flask(__name__)
app.config['SQLALCHEMY_DATABASE_URI'] = os.environ['SQLALCHEMY_DATABASE_URI']
app.secret_key = "supersecretstring"
db = SQLAlchemy(app)
langchain_memory = LangchainFlaskMemory(app, db)
llm = OpenAI(temperature=0)
@app.route('/')
def index():
langchain_memory.set_user_id("abc123")
return "Hello, World!"
@app.route('/chat', methods=['POST'])
def chat():
input = request.json['message']
m = langchain_memory._create_chat_memory()
c = ConversationFlaskMemory(chat_memory=m, return_messages=True)
conversation = ConversationChain(
llm=llm,
verbose=True,
memory=c,
)
answer = conversation.predict(input=input)
return jsonify({"message": answer})
@app.route('/add')
def add():
chroma = langchain_memory._create_chroma_vector_store()
chroma.add_texts(texts=["doc1", "doc2", "doc3"])
return "Added texts"
@app.route('/count')
def count():
chroma = langchain_memory._create_chroma_vector_store()
return str(chroma._collection.count())
if __name__ == '__main__':
app.run(debug=True)