rain forge
There're a number of frustrations with Langchain I share (https://www.reddit.com/r/LangChain/comments/13fcw36/langchain_is_pointless/). New & better chat models, vector db products, and vastly bigger context sizes leave parts of the library looking fairly over-engineered.
I built GPTIO (https://github.com/gsidsid/gptio) to give myself a simple interface I understood to let GPT-4 reason through tasks and execute custom JS functions with multiple inputs, a feature that unbelievably was just merged today into Langchain– and one which still appears poorly implemented (StructuredTool class + Zod schemas) and poorly documented (only documentation here?? -> https://github.com/hwchase17/langchain/blob/master/docs/modules/agents/tools/multi_input_tool.ipynb).
npm i gptio to get started. A code sample is available on the readme that should contain everything you need. Long term memory can easily be introduced with whatever data source you need by writing a retrieval action of your own.
EDIT: now features memory, integrates with every single vector database https://github.com/gsidsid/gptio#memory
