OpenGPTs is a take on what an open-source GPT store may look like, which was launched two months ago and updated recently to use LangGraph and added new features. It runs on MessageGraph, a particular type of graph that is special in that each node takes in a list of messages and returns messages to append to the list of messages. OpenGPTs supports three different cognitive architectures: Assistants, RAG, and ChatBot, which can be used by users when creating bots. The Assistants are the most flexible choice but also work well with fewer models and can be less reliable. Persistence is baked in OpenGPTs via LangGraph checkpoints, allowing for the surface of messages of past conversations. Configuration is achieved through LangChain primitives, making it easy to save and load configuration options. New models have been integrated, including Google's Gemini model, and new tools have been added, such as Robocorp's Action Server, which can be used to define arbitrary Python functions as tools. The astream_events method is used to stream back events and surface them to the user.