PsyFiGPT Now Reads Your PsyFi Scribe Sessions: One AI, Two Surfaces

Quick answer: PsyFiGPT can now pull a completed PsyFi Scribe session transcript into a chat conversation via the new get_scribe_transcript @tool. Ask "Summarize my 2 PM session" or "Draft a treatment plan update from today's session" and PsyFiGPT fetches the transcript on demand, with PHI staying inside the PsyFi-controlled environment. No copy-paste, no manual export — one BAA covers both surfaces.

The problem: clinicians juggle one tool for recording, another for follow-up reasoning

A typical post-session workflow looks like this. You finish the session. PsyFi Scribe (or whatever ambient documentation tool you use) has the transcript. You open a separate chat window — sometimes with a HIPAA-aligned product, sometimes, less safely, with whatever consumer chatbot is open — and try to summarize, draft a treatment plan update, or generate a referral letter.

Usually you are the bridge between the two tools. You copy the transcript out of one system. You paste it into another. You either hope the second system is BAA-backed, or you redact on the fly and lose the texture that made the transcript useful.

Every handoff is a workflow tax, and every handoff between two clinical surfaces is a PHI risk. Even when both tools are HIPAA-aligned, moving the transcript through your clipboard expands the surface area of where that data has lived. Audit trails fragment. Retention windows get harder to reason about. And if you ever paste into a non-BAA tool by mistake, you have a breach-notification calculus to run.

This is the gap the new integration closes. PsyFiGPT can now ask PsyFi Scribe for the transcript directly — no clipboard, no export, no second system involved.

How the integration works

The mechanism is the @tool framework that shipped to PsyFiGPT earlier this year. Tools are scoped functions the AI can invoke mid-conversation when a request needs information it does not already have. get_scribe_transcript is the newest of these, wired specifically to the Scribe backend.

The end-to-end flow has three steps:

1. PsyFi Scribe records the session. The transcript is finalized and stored inside PsyFi's HIPAA-aligned infrastructure, exactly as it always has been.
2. You open a PsyFiGPT chat and ask a question that requires session content — "Draft a SOAP note from this morning's session," "What did we agree on as homework for the patient I saw at 2 PM?", "Pull the transcript from the third session this week."
3. PsyFiGPT calls get_scribe_transcript. The tool resolves the request — either by session ID if you provided one, or by recency and rough time window if you described it conversationally — and returns the transcript text into the AI's working context. PsyFiGPT then answers using that content.

Every call is logged. The same audit trail that records who viewed a session inside PsyFi Scribe records who pulled it through PsyFiGPT, when, and which chat conversation requested it. This matters because a tool that touches PHI without an audit trail is not really a HIPAA-aligned tool — it is a future incident report.

Three example workflows

Example 1: Drafting a treatment plan update

You ended a session this morning where the patient described a meaningful shift — work-related anxiety has receded as a new pattern of family conflict has emerged. The treatment plan you wrote 90 days ago does not reflect this.

Open a PsyFiGPT chat. Ask: "Pull this morning's 9 AM session and draft a treatment plan update reflecting any changes in presenting concern." PsyFiGPT calls get_scribe_transcript, reads the session, and returns a draft that highlights the shift, suggests an adjustment to the treatment focus, and flags the goal language that needs revision. You review, edit, and paste into your EHR.

The takeaway: the stale treatment plan problem is partly a documentation problem and partly a context problem. Pulling the transcript directly into the drafting environment closes both gaps at once.

Example 2: Finding patterns across sessions

A patient has been in treatment for four months. You want to step back and ask whether the themes you have been treating are still the central concerns, or whether the work has drifted.

Ask PsyFiGPT: "Pull the last six session transcripts for this patient and tell me what themes recurred, which goals were referenced, and where the conversation moved." PsyFiGPT calls get_scribe_transcript for each session, reads them, and returns a thematic summary that you can use as a clinical reflection tool before your next supervision meeting or treatment review.

The takeaway: cross-session pattern analysis is one of the highest-leverage uses of AI in behavioral health, and one of the most awkward to do manually. The integration makes it routine instead of aspirational.

Example 3: Generating a referral letter using session observations

A patient needs to be referred to a psychiatrist for a medication evaluation. You want the referral letter to include specific observations from recent sessions, not generic boilerplate.

Ask PsyFiGPT: "Pull the last two session transcripts and draft a referral letter to Dr. X requesting a medication consult. Include specific examples of symptoms the patient described and any functional impact they reported." PsyFiGPT pulls both transcripts, drafts a letter that quotes or paraphrases the relevant observations, and returns it for your review.

The takeaway: referral letters that include concrete clinical detail are more useful to the receiving clinician and stronger for medical-necessity documentation. The bottleneck has historically been the time it takes to mine recent transcripts for the right details. That bottleneck is now an AI call.

HIPAA architecture

The cross-product nature of this feature is exactly the kind of workflow that raises HIPAA questions, so it is worth being precise about what is and is not happening.

Same BAA. Your existing Business Associate Agreement with PsyFi covers both PsyFiGPT and PsyFi Scribe. The integration introduces no new vendor relationship and no new contract to review. If your practice has signed a BAA with PsyFi, the get_scribe_transcript tool is already covered.

Same data residency. The transcript never leaves PsyFi's infrastructure. The tool call is a server-to-server fetch inside PsyFi's environment. The transcript is loaded into PsyFiGPT's working context only for the duration of the conversation that requested it.

Same audit log. Every get_scribe_transcript call is logged with the requesting user, the session it pulled, the timestamp, and the PsyFiGPT conversation ID. Practice owners can review these logs the same way they review any other access to a Scribe session.

No PHI to third-party LLM providers. The tool framework was designed so PHI does not pass through any third-party LLM provider's logging or training pipeline. The model sees the transcript as conversation context, but the inference path keeps that content inside the BAA-covered environment.

This is part of our complete guide to HIPAA-compliant AI for behavioral health practices.

What this replaces

For practices already using PsyFi Scribe and PsyFiGPT, this integration replaces a workflow that, before today, had no good options.

Copy-paste between two HIPAA-aligned tools. Even when both tools were BAA-backed, the act of moving a transcript through the clipboard expanded the data's footprint and broke the unified audit trail. Practice owners had to reconstruct who accessed what by reconciling two separate logs. With the integration, the access is a single auditable event.

Manual transcript review for post-session reasoning. Reading a 50-minute transcript to find the two clinically important moments is not a good use of clinician time. AI is faster at this kind of pattern-matching, and the integration makes that speed available without the data-movement risk.

Exposure to consumer ChatGPT. This is the workflow nobody likes to talk about, but it is the most common one this integration replaces. Clinicians without a HIPAA-aligned reasoning tool reach for whatever is open, and consumer ChatGPT is usually open. See Is ChatGPT HIPAA Compliant? 2026 Guide for Behavioral Health Practices for what NOT to do, and why the architectural difference matters more than the marketing copy on any vendor page.

The integration is not a cure for every cross-product workflow. It is the first time that "reason about a recorded session in a HIPAA-aligned AI chat" is a single, auditable, BAA-covered action rather than a sequence of moves the clinician has to choreograph.

How to enable it

If your practice account has both PsyFiGPT and PsyFi Scribe, the tool is already available. There is no toggle to flip and no setting to configure. The next time you open a PsyFiGPT chat and ask a question that requires session content, the AI will call get_scribe_transcript on its own.

A few practical notes for getting reliable results:

• Be specific when you can. "Pull the session from this morning at 9 AM" gives the tool a clean recency target. "Pull last week's session with the patient who mentioned the custody hearing" works less reliably because the tool is matching on time window, not transcript content.
• Sessions older than your retention window are not available. Default retention is 90 days. If you ask for a session from before that window, the tool returns a "not found" response and the AI will tell you.
• Audit logs are visible to practice administrators. If your practice has compliance review built into its operations, the get_scribe_transcript activity is reviewable the same way other Scribe access is.

If you want to understand how PsyFiGPT picks and runs tools mid-conversation, see our explainer on the @tool framework. For a broader view of why purpose-built behavioral-health AI matters, PsyFiGPT and PsyFi Scribe are designed to work together precisely because most clinical workflows do.

This post is for informational purposes only and does not constitute legal advice. Consult a healthcare attorney for guidance specific to your practice's compliance obligations.