An Innovative Approach to Instantly Adding Memory Capabilities to OpenAI API Applications

[narrowizard]

It looks like Letta has implemented a fully functional agent that supports long-term memory and provides a set of APIs for applications. I suddenly had an idea – if Letta could implement an API compatible with OpenAI, acting as an OpenAI proxy, and automatically enabling memory functions during proxy requests (and responses), then existing applications based on the OpenAI API could immediately have memory capabilities. (The only change needed would be to modify the API base URL to point to Letta's proxy API, without modifying any code).

[narrowizard]

A more practical question is, how can I quickly migrate an application developed based on the OpenAI API to Letta?

[sarahwooders]

We have considered adding an OpenAI ChatCompletitions proxy to our API - the one caveat is that you would need to provide an additional agent_id field into the request, so that we know which agent that chat request should go to. It's also a bit weird since unlike ChatCompletitions, you should only be providing new messages, rather than the full message history.

Our current API is designed to be more suited towards a stateful agents model where the developer does not need to manage the chat history https://docs.letta.com/api-reference/overview -- our messages endpoint is very similar to ChatCompletitions https://docs.letta.com/api-reference/agents/messages/create, but we'd be open to supporting a fully compatible endpoint if there are strong reasons to do so.