We are utilizing the open source Llama3.1 405B model from Meta, which has shown impressive benchmark results compared to cloud-based alternatives. You are right, some AI models have their own built-in safeguard layers. You can try some Anon agent without built-in safeguard.