We are currently cutting the prompt context at 8000 characters, we need a better strategy. Options are: - Using embeddings - Make multiple calls to chatGPT API