AI Scribe
Praxi's AI Scribe transforms your session recordings into structured clinical notes while preserving your own handwritten notes. It combines voice transcription, AI generation, and smart merge to produce high-quality, accurate clinical documentation in seconds.
How It Works
The AI Scribe integrates into Praxi's three-phase session workflow:
| Phase | What Happens |
|---|---|
| Prepare | Test your microphone, review pre-session brief, check consent status |
| Active | Record the session, write notes alongside, view live session markers and clinical guidance |
| Wrap Up | AI generates a note from the transcript, merges with your handwritten notes, review and sign |
The Fast Path (3 Clicks)
For practitioners who want speed:
- End session — AI auto-generates a note using your default template
- Scan the merged note — your handwritten content stays at the top of each section, AI content appears below with a subtle tint
- Sign — the note is locked and audit-trailed
No extra clicks needed. The default flow is designed for busy clinicians.
Smart Merge
The AI Scribe never overwrites your handwritten notes. When you write notes during a session and then generate an AI note from the transcript:
- Your content stays at the top of each section
- AI content appends below a divider, marked with a subtle background tint
- Empty sections are filled by the AI directly
- An optional Review AI Merge button lets you choose per section: Keep Mine, Use AI, or Merge Both
AI-generated content is visually distinguished with a light tint. Typing within an AI-marked span automatically clears the mark, so your edits blend naturally.
Verification and Trust
Verification Flags
The AI flags claims it cannot trace to the transcript with inline [VERIFY]
badges. Click a badge to see why it was flagged and choose to:
- Confirm — you've verified it's accurate, badge is removed
- Edit — modify the text inline
- Remove — delete the unverified claim
Unresolved flags trigger a warning before signing, but never block it.
Source Citations
AI-generated sentences include small superscript citation numbers linking back to the transcript segment that supports each statement. Hover to see the quoted transcript text and timestamp. Click to open the transcript drawer scrolled to that moment.
Confidence Score
An overall confidence percentage appears in the session summary (green >85%, amber 70-85%, red <70%) so you can gauge how much review is needed.
Templates and Customisation
Note Templates
Choose from built-in templates:
| Template | Sections | Best For |
|---|---|---|
| SOAP | Subjective, Objective, Assessment, Plan | General clinical notes |
| DAP | Data, Assessment, Plan | Streamlined documentation |
| BIRP | Behaviour, Intervention, Response, Plan | Behavioural focus |
| GIRP | Goals, Intervention, Response, Plan | Goal-oriented therapy |
| Narrative | Free-form clinical documentation | Flexible approach |
| ADIME | Assessment, Diagnosis, Intervention, Monitoring, Evaluation | Dietetics |
| ICF | International Classification of Functioning framework | Functional focus |
| Progress Note | Summary, Observations, Plan | Quick progress updates |
40+ profession-specific templates are also available, including specialised formats such as EMDR protocol notes for trauma-focused therapy, across psychology, counselling, physiotherapy, occupational therapy, speech pathology, dietetics, and more.
Smart Sections
In Settings > Note Templates, customise individual sections of your templates:
- AI instructions — tell the AI what to focus on per section (e.g., "Always include PHQ-9 scores when available")
- AI toggle — disable AI generation for sections you always write manually
- Both settings are saved to the template and apply to all future generations
Style Learning
The AI learns from your edits. After signing an AI-generated note, Praxi analyses the differences between the original AI output and your final version. Over time, generated notes progressively match your terminology, tone, section lengths, and formatting preferences.
This happens automatically. No configuration needed.
Context Sources
The AI uses multiple context sources to generate accurate notes. In the generation panel, you can toggle each source:
| Context | Description | Default |
|---|---|---|
| Full Transcript | The complete session recording transcript | On |
| Recent Notes | Last 3 clinical notes for continuity | On |
| Assessment Results | Recent outcome measure scores (PHQ-9, GAD-7, K10, etc.) | On |
| Treatment Plan | Active treatment goals and interventions | On |
| Uploaded Files | Referral letters, intake forms, external reports (max 5) | Off |
More context improves note quality but increases generation time and cost. An estimated cost and duration is shown in the generation panel.
Preferences
Configure your AI Scribe defaults in Settings > AI Preferences > Clinical Note Defaults:
| Preference | Options |
|---|---|
| Default template | SOAP, DAP, BIRP, Progress Note |
| Detail level | Brief, Moderate, Comprehensive |
| Person reference | "The client", "The patient", "They/them", Client's name |
| Quality vs speed | Fast, Balanced, or Best Quality (cost varies by note length; the picker shows a live estimate). Available options depend on your subscription tier |
| Auto-generate | Automatically start AI generation when session ends |
| Show AI highlighting | Visually distinguish AI content in the editor |
Golden Thread
After generating a note, the Golden Thread panel in the wrap-up sidebar shows how the current session connects to the treatment arc:
- Presenting problem
- Active diagnoses
- Treatment goals (with progress from this session)
- Interventions used
- Outcome measures
- Next session focus
This ensures every session note demonstrably connects to the treatment plan (AHPRA best practice).
MBS Code Suggestions
Based on session type, duration, and content, the AI suggests applicable Medicare item numbers as clickable badges. This streamlines billing by connecting notes directly to MBS claims.
Live Session Intelligence
During active recording (PRO tier and above; full guidance for telehealth and phone sessions, risk alerts only for in-person and couples sessions):
- Session markers appear as lightweight tags on a timeline: "Topic: childhood experiences" at 12:34, "Intervention: CBT thought challenging" at 18:02
- Live guidance from Praxi shows reflective questions and reminders from your documented plan (telehealth and phone sessions only)
Recording Setup
Before recording, use the Microphone Setup card in the Prepare phase to:
- Select your preferred audio input device
- Run a 3-second mic test with playback
- Verify audio quality
The selected device is remembered for future sessions.
Consent
Voice transcription requires a two-layer consent process:
- Broad consent — recorded once per client (valid 12 months, auto-renewal prompt)
- Per-session confirmation — practitioner confirms consent is current at session start (auto-skipped for fresh consent via smart gate)
AI processing of clinical data requires separate aiProcessingConsent.
Compliance
| Requirement | How Praxi Handles It |
|---|---|
| AHPRA contemporaneous documentation | 24-hour signing deadline (configurable per organisation) |
| AI disclosure | Optional disclosure footer: "This note was generated with AI assistance and reviewed by the signing practitioner" |
| AI attestation | Practitioner must review and attest before signing AI-generated notes |
| Privacy Act 1988 | Audio not kept by default; a practice can keep it up to N hours for playback. Transcript retained 7 days after signing, consent required |
| Audit trail | 7-year immutable trail for all note operations |
| Risk documentation | Risk flags displayed inline, HIGH/CRITICAL require acknowledgment before signing |
Subscription Tiers
| Feature | STUDENT | ESSENTIALS | PRO | ENTERPRISE |
|---|---|---|---|---|
| Batch transcription | Yes | Yes | Yes | Yes |
| AI note generation | Yes | Yes | Yes | Yes |
| Haiku model | Yes | Yes | Yes | Yes |
| Sonnet model | -- | Yes | Yes | Yes |
| Opus model | -- | -- | -- | Yes |
| Live clinical guidance | -- | Yes | Yes | Yes |
| Session memory | -- | Yes | Yes | Yes |
| Multi-speaker (couples/group) | -- | -- | Yes | Yes |
| Growth analytics | -- | -- | Yes | Yes |
Troubleshooting
| Issue | Solution |
|---|---|
| Mic not detected | Check browser permissions (Settings > Privacy > Microphone) |
| Low confidence score | Try including more context sources or using a higher quality model |
| [VERIFY] flags on accurate content | Confirm the flag to remove it — the AI is being cautious |
| Wrong speaker labels | Use the speaker swap controls in the transcript viewer |
| Generation failed | Check your monthly AI credit balance in Settings > AI |