Moving from a consumer AI tool to a HIPAA compliant one is mostly about recreating your setup, not relearning how to work. Here is what transfers, what does not, and the fastest path.
What does not transfer
There is no direct import from ChatGPT: chat history, custom GPTs, and memory cannot be pulled in automatically, and for privacy reasons we would not want account-to-account access anyway. Note that OpenAI lets you export your ChatGPT data from its settings if you want a personal archive before you switch.
Recreate your custom GPTs as saved prompts
BastionGPT's equivalent of a custom GPT is a saved prompt: reusable instructions plus attached reference documents, stored in the Prompt Library and invoked by typing "/". Copy the instructions out of your GPT, paste them into a new saved prompt, attach the same reference files, and you are back to one-keystroke workflows. Saved prompts can also be shared with colleagues, which is how teams standardize documentation styles.
What works differently, on purpose
No web search. Responses come from the models and your documents, not live browsing. See why BastionGPT doesn't have web search.
No cross-chat memory. Each chat starts fresh; saved prompts are how you carry context forward deliberately rather than accidentally.
Same frontier models. BastionGPT routes across the latest models from OpenAI, Anthropic, and Google, so response quality will feel familiar; the difference is the compliance wrapper around them.
And the reason you are switching in the first place: every plan includes the HIPAA BAA, your data is never used to train models, and patient information stays inside a compliant environment, so the copy-paste-into-ChatGPT risk disappears from your practice.