PsyFi Scribe captures behavioral health intake sessions on-device — CC, HPI, MSE, treatment-plan starter. Built for 1-4 speaker sessions (solo, family, interpreter). Group therapy is out of scope, and we'll tell you why.
Quick answer: PsyFi Scribe captures behavioral health intake sessions on-device and produces structured intake summaries: chief complaint, history of present illness, mental-status observations, and a treatment-plan starter. Built for 1–4 speaker sessions, it handles the common intake configurations — solo intake, intake with a parent or partner, intake with an interpreter — without requiring you to identify speakers manually. Group therapy is out of scope.
An intake session is 45 to 60 minutes of dense, clinically critical content. The patient is laying out their reasons for seeking treatment, their medical history, their current symptoms, what they've already tried, and the context of their life right now. Every detail matters. Everything downstream depends on getting this right — your diagnosis, your treatment plan, your risk assessment, the phrasing you use in insurance authorizations.
And the documentation burden is enormous. A typical clinician's approach is to take notes during the session (which diverts attention from the patient), then reconstruct the narrative from memory and fragments afterward. The result is often a hybrid of what was said, what you remember, what you guessed at, and what seemed clinically important. Gaps form. Details fall away. By the time you're writing the treatment plan two days later, some of what the patient told you is already half-lost.
This is exactly the moment where an ambient documentation tool helps. The session is recorded, the transcript is available, and the structuring work — turning 60 minutes of unstructured conversation into a formal intake note — becomes tractable. Not invisible, but no longer manual reconstruction from fragments.
The stakes are also highest here. Intake is the peak PHI moment. A patient's first contact with your practice brings name, address, insurance, reason for treatment, medical history, trauma history, substance use, suicidality, and personal details that they've just disclosed to someone they've known for an hour. The data density is higher than almost any later session, and the clinical consequence of getting the intake wrong is steeper. So the compliance burden is steeper too. SAMHSA guidance on initial assessment emphasizes that every intake must establish baseline diagnosis, risk, and treatment goals. Documentation that skips those is not just incomplete — it's clinically and legally insufficient.
PsyFi Scribe exists partly because intake is where an automated-recording approach pays the highest dividends.
An intake note is not a therapy progress note. It is a comprehensive clinical assessment designed to establish the patient's baseline and chart the direction of treatment. The typical structure includes five core sections.
Chief Complaint (CC) — the patient's stated reason for seeking treatment in their own words. Not your interpretation, not your clinical paraphrase: what the patient said. Example: "I've been anxious for months and it's affecting my work." This is the hook that everything else hangs on.
History of Present Illness (HPI) — the timeline and trajectory of the chief complaint. When did it start? What triggered it? How has it changed? What have they tried? What makes it better or worse? The HPI is where you build the narrative of the symptom constellation and establish which symptoms are primary and which are secondary.
Mental Status Exam (MSE) — your clinical observations of the patient during the intake: appearance (grooming, age presentation), behavior (motor activity, eye contact, cooperativeness), mood (what they report feeling), affect (what you observe), thought process (how they organize their thoughts), thought content (what they're thinking about — especially any red flags like suicidality or grandiosity), cognition (orientation, memory, attention), and insight/judgment (how well they understand their condition and can plan).
Risk Assessment — screening for suicide, self-harm, harm to others, substance use, homelessness, and other acute safety factors. These are usually yes/no flags or severity ratings, often captured in a structured format so you can quickly scan them later.
Treatment Plan Starter — your initial clinical impressions feeding into a preliminary plan. Working diagnosis (or diagnostic possibilities if you're not certain yet), treatment modality (individual therapy, medication consult, etc.), frequency, and initial goals. The formal treatment plan usually comes later, after you've had time to reflect, but the intake note should contain the seeds of direction.
The APA's guidelines on initial assessment and standard practice by most licensing boards all require these elements in some form. An intake note that lacks them is incomplete.
The content of an intake session is structurally different from an ongoing therapy session, and an AI scribe has to account for that.
In an ongoing session, the clinician is directing: "Tell me more about that." "What happened next?" "How did that land for you?" The session is typically higher in clinician talk — 40 to 50 percent clinician, 50 to 60 percent patient. The patient is responding to the clinician's questions.
In an intake, the clinician is gathering: "Tell me about your family." "When did the anxiety start?" "What was your childhood like?" The patient is doing the bulk of the talking — 60 to 70 percent patient, 30 to 40 percent clinician. It is more open-ended, more narrative-driven. The clinician is asking broad questions and the patient is filling in a timeline and history.
This matters for diarization (correctly labeling who said what) and summary generation. An AI scribe optimized for ongoing therapy has to shift its approach for intake. It has to capture the patient's narrative arc, not just the question-answer exchanges. It has to extract the timeline and trajectory, not just identify shifts in clinical focus.
PsyFi Scribe was designed with intake in mind from the start. It captures the chronology, the self-reported symptoms, and the patient's framing of the problem — and it distinguishes those from the clinician's observations and assessment.
The workflow is straightforward: record, diarize, summarize, hand off.
Step 1: Record. You open PsyFi Scribe on your device (phone, tablet, or laptop) at the start of the intake session. The app begins capturing audio on-device — that is, the audio file is stored locally on your device and never transmitted to a cloud server during the session. You can conduct the intake exactly as you normally would. At the end of the session, you tell Scribe the session is complete.
Step 2: Diarize. Scribe processes the recording and identifies each distinct speaker. If it's a solo intake (clinician + patient), it identifies two speakers. If a parent joins, it identifies three. If there's an interpreter, it's still three: clinician, patient, interpreter — each recognized as a separate voice. Scribe labels each segment of the transcript with the speaker: "Patient: I've been feeling..." "Clinician: And how long has that been going on?" The diarization happens on your device or is completed within PsyFi's secure infrastructure, but PHI stays within the PsyFi environment.
Step 3: Summarize. Once diarization is complete, Scribe generates a structured intake summary. It extracts the chief complaint, identifies the timeline and history, pulls out clinical observations, notes risk indicators, and proposes a treatment-plan starter. The summary is organized in the standard sections: CC, HPI, MSE, risk, and treatment plan starter. All of this is available to you in your PsyFi account.
Step 4: Hand off. The transcript and summary become available in your PsyFi Scribe account. From there, you can export them, use them to draft your formal intake note in your EHR, or — if you also use PsyFiGPT — pull the transcript directly into a chat conversation to draft treatment plans, find clinical patterns, or generate a referral letter from the intake. No copy-paste. The transcript lives in one secure place and you reference it as needed.
PsyFi Scribe's diarization model is optimized for clinical recording environments with up to 4 distinct voices. This is not arbitrary — it's a choice based on model accuracy and real-world clinical workflows.
Solo intake — clinician + patient. This is 2 speakers and works flawlessly. Scribe correctly attributes every statement. The patient describes their symptoms, the clinician asks clarifying questions, the transcript is crisp and accurate.
Family intake — clinician + patient + 1 or 2 family members (parent, partner, adult child). This is 3 or 4 speakers. Scribe diarizes each voice and correctly attributes statements. This matters clinically because collateral information (what a family member observes about the patient) is often different from the patient's self-report and both belong in the note. Scribe captures that distinction.
Intake with an interpreter — clinician + patient + interpreter. This is 3 speakers. Scribe recognizes each voice. The interpreter is captured separately from the clinician and patient, so you have a clear record of what the patient said (in their language) and what the interpreter conveyed. This is especially important for complex intake assessments where the nuance of the patient's expression matters.
Large family meetings or group settings — clinician + 5 or more speakers. This is out of scope. When you exceed 4 speakers in a recording, speaker overlap increases, acoustic complexity grows, and diarization accuracy degrades sharply. Voices get misattributed, overlaps become garbled, and the summary becomes unreliable. We chose to do the 2-4 speaker case really well rather than pretend to handle 6+ speakers and fail at both.
This is a hard constraint and it is intentional. Group therapy, classroom settings, large workshops — those are structurally outside the model's reliable range. If you are running intakes that routinely include 6+ voices, PsyFi Scribe is not the right tool. There are other documentation approaches for group settings; we are not building one.
Intake is the highest-PHI moment in the patient relationship and it deserves corresponding compliance attention.
Consent first. When do you obtain consent for AI recording? Before the session starts. Update your intake consent form to include a line: "This session will be recorded with artificial intelligence-assisted transcription for clinical documentation purposes. Your information is not used to train AI models and is covered by our Business Associate Agreement with PsyFi." Have the patient sign or acknowledge before you start the recording. Document it in the patient's record.
Scope of the BAA. The Business Associate Agreement that covers PsyFiGPT also covers PsyFi Scribe. If your practice has signed a BAA with PsyFi, both the recording and the AI-generated summary are covered. No additional contract is needed. Same vendor, same legal protection, same audit log.
Data residency. The recording and transcript stay inside PsyFi's HIPAA-aligned infrastructure. They are not sent to a third-party cloud LLM provider. They are not used for model training. They are not shared with other patients' clinicians. PsyFi operates a per-tenant data model for each practice, which means your data is isolated from other customers' data by default.
Retention defaults. PsyFi Scribe's default retention is 90 days. After 90 days, recordings and transcripts are automatically deleted unless you tag them for permanent retention. You can request shorter retention (e.g., 30 days) or longer (e.g., 7 years to match your medical record retention). This is configurable per practice and documented in your account settings.
This is part of our complete guide to HIPAA-compliant AI for behavioral health practices. For the federal-rule deep dive, see Is ChatGPT HIPAA Compliant?. For specifics on intake consent and data handling, see Is AI-Assisted Intake HIPAA Compliant? and Data Retention and Security in AI-Assisted Intake.
This is the hard truth and it is important to state clearly: PsyFi Scribe is not built for group therapy intakes and it will not work well for large family meetings with 5 or more speakers.
The 4-speaker limit is not a soft preference — it is the boundary of where the diarization model performs reliably. A group therapy intake with 8 voices will produce a transcript where:
This is the kind of trade-off where doing it badly is worse than not doing it at all. We chose to optimize for the 95 percent of intakes that are 1-4 speakers and explicitly not serve the 5 percent that require larger-group handling.
If you run group intakes regularly, you need a different documentation approach. There are other tools designed for group settings, though they typically involve more manual effort or live transcription with human review. We are not building one.
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.