BastionGPT does not carry an overarching memory from one chat to the next. This is a deliberate design choice rooted in patient safety and HIPAA compliance. In a healthcare AI assistant, any mechanism that automatically transfers context between sessions creates a pathway for protected health information belonging to one patient to surface in a conversation about another. For clinicians juggling dozens of encounters a day, that kind of cross-contamination is a documentation and privacy hazard inappropriate to introduce.
The underlying issue is accuracy. Filtering sensitive details out of a shared memory store would need to work flawlessly every single time, across every specialty, note type, and edge case. Current techniques do not reach that threshold. Until memory systems can demonstrate complete reliability in scrubbing identifiers and preventing patient data from crossing session boundaries, BastionGPT will continue to treat each chat as its own isolated workspace. This is part of why healthcare professionals, from physicians to therapists writing progress notes, trust the platform as a top choice for secure clinical documentation.
That said, we recognize clinicians do not want to retype the same background information every time they start a new session. To solve that without compromising privacy, BastionGPT offers a saved prompt feature. You can store reusable instructions, templates, formatting preferences, specialty-specific guidance, or standing context relevant to particular workflows, and drop them into any new chat with a click. Think of it as a personal library of building blocks for your AI medical scribe or therapy notes assistant, fully under your control and free of patient identifiers.
This approach gives you the efficiency benefits of continuity without the compliance exposure of persistent memory. A psychiatrist can keep a saved prompt for structuring mental health progress notes. An IT administrator can standardize prompts across a clinical team. A clinical supervisor can store teaching frameworks for case discussions. In each scenario, the user decides exactly what information enters a conversation, and nothing carries over implicitly from prior sessions.
What makes AI memory risky with patient data?
Cross-chat memory works by learning from your conversations. When those conversations contain patient data, that means training an AI on protected health information, and it creates two risks we are not willing to accept:
Consent. With memory, one patient's information helps shape what the AI writes about another patient. Unless your consent forms specifically cover that, your patients have not agreed to it, which creates regulatory and consent problems for your practice.
Memory leak. An AI that learns from patient data can repeat it back later. When two patients look similar on paper, a memory-enabled AI can confuse them and place one patient's details in another patient's note. That is a potential HIPAA violation, and a FERPA violation in school settings.
This reflects a rule we apply across BastionGPT: we will never ship a feature that could put your compliance at risk. Anywhere data can go, we assume patient data will go, so a feature has to be safe for PHI before it ships. Your data is also never used to train or refine AI models, and we guarantee that contractually.
How do I get continuity without cross-chat memory?
Every chat starts fresh, and you control exactly what context it starts with:
Add templates and writing samples to saved prompts. Alongside your instructions, attach your report template and one or two samples of your ideal writing. The AI will match your structure, tone, and style in any new chat. In our experience, instructions plus a real sample works better than instructions alone.
Set standing instructions. Save the preferences you want applied every time, such as terminology standards, formatting rules, and tone. Teams can share prompts so the whole organization documents consistently.
Attach prior documents to the chat. If the AI needs a treatment plan, prior report, or clinical history, attach the file or paste the text into the new chat. You decide, patient by patient, exactly what the AI sees.
Organize chats into folders. Create a folder per client so earlier chats and their outputs are easy to find and refer back to.
Repeat critical instructions in long prompts. In long prompts and extended conversations, state your most important instructions more than once, at different points. Even leading AI companies repeat key instructions several times within their own prompts because models follow them more reliably that way.
For step-by-step setup, see saved prompts in the Prompt Library. For ideas on what to store, read our guide to building a prompt library.
Can BastionGPT reference a client's treatment plan when I write the next session note?
Yes, with one quick step. BastionGPT will not automatically recall the treatment plan from an earlier chat, but you can attach the document (or paste the relevant text) at the start of the new session and ask the AI to reference it while drafting your note. Many clinicians keep a per-client folder of chats so those materials are easy to find for each session.
If those notes build toward a lengthy document, see our guide to generating long reports without losing quality.
If and when memory technology matures to the point where sensitive information can be filtered with complete certainty, we will revisit the feature. Until then, session isolation combined with saved prompts reflects what a HIPAA compliant AI platform should look like: powerful enough to accelerate clinical work, disciplined enough to keep every patient's information exactly where it belongs.