> Maybe this has long been obvious to everyone but me, but having cloned a few repositories of ChatGPT API web UIs, the API itself really is dead simple: send y

> Maybe this has long been obvious to everyone but me, but having cloned a few repositories of ChatGPT API web UIs, the API itself really is dead simple: send your prompt and then with every subsequent include the entire conversation history. The streaming of the response is the most novel aspect of the actual API call. It pretty soon gets complex if you are looking to update the prompt (which would require a code change or re-deployment, understand how the changes to the prompt affect the outcome, caching etc. And that complexity increases with number of prompts you are managing. In addition, another thing we observed is that in large scale projects, people who add these prompts would be some sort of subject experts that are different from people who manage the code.