AI Settings & Permissions
Praxi's AI features are configurable at the platform, organisation, and practitioner level. This guide covers enabling and disabling AI agents, model tiers, approval requirements, consent, and audit trails.
Enabling and Disabling AI Agents
Praxi includes six specialised AI agents:
| Agent | Category | Domain |
|---|---|---|
| Praxi | Clinical | Clinical notes, reports, compliance checks |
| Jarvis | Practice Operations | Scheduling, billing, Medicare, help, resource booking |
| Ada | Knowledge Management | Research, evidence, MBS guidance, knowledge curation |
| Warren | Business Strategy | Analytics, revenue, marketing, growth insights |
| Brene | People Management | Staff wellness, HR, team dynamics, onboarding |
| Linus | Technical | Workflows, automation, integrations, website, reports |
To enable or disable specific agents:
- Navigate to Settings > AI Preferences
- Toggle each agent on or off
- Save your changes
Disabling an agent prevents it from being invoked by any user in your practice. Specialist agents (the clinical assistant, Ada, Warren, Brene and Linus) require a Pro plan or higher -- on Student and Essentials plans only Jarvis and the Praxi orchestrator are available, regardless of toggles.
Cost Controls
Praxi provides daily and monthly cost limits to help you manage AI spending:
- Navigate to Settings > AI
- Set your Daily limit (minimum $1, maximum $1,000)
- Set your Monthly limit (minimum $10, maximum $10,000)
- Choose what happens when a limit is reached:
| Option | Behaviour |
|---|---|
| Notify | Send a notification but continue processing |
| Pause | Pause AI features until the limit resets |
| Downgrade model | Switch to a more cost-efficient model automatically |
| Block | Stop all AI processing until the limit resets |
The AI panel shows a budget status indicator when spending approaches your limits. You can view current usage (daily and monthly spend, request counts) in Settings > AI.
any time. The system warns you at 80% of your limit. If your practice has overage billing enabled, AI usage can continue beyond your limits up to a monthly overage cap rather than pausing or blocking. :::
Model Tier by Subscription
The AI model available to your practice depends on your subscription tier:
| Tier | Maximum Model | Agents Available |
|---|---|---|
| Student | Haiku (fast, efficient) | Jarvis + Praxi orchestrator |
| Essentials | Haiku (fast, efficient) | Jarvis + Praxi orchestrator |
| Pro | Sonnet (balanced) | All six agents |
| Enterprise | Opus (most capable) | All six agents |
model upgrades within your tier ceiling. For example, a high-risk clinical query on the Team plan may escalate from the efficient model to the balanced model.
Evidence Sources ("What Ada reads")
The What Ada reads section of Settings > AI controls which evidence databases and source tiers AI agents draw from for research and evidence search. Choose a system preset or build a custom preference — see Evidence Search for the full options.
Data Access Controls
Control what data AI agents can access:
- Navigate to Settings > AI
- Configure the following toggles:
| Setting | Description | Default |
|---|---|---|
| Allow client data | Let AI access client records for context | On |
| Allow clinical notes | Let AI read and assist with clinical notes | Off |
| Allow voice transcription | Enable AI-powered voice transcription features | On |
| Require client consent | Require recorded client consent before AI processes data | On |
only after reviewing your practice's privacy and consent policies. :::
Per-Agent Configuration
Each agent can be individually configured:
- Navigate to Settings > AI Preferences
- Click on an agent to configure it
- Available settings:
| Setting | Description |
|---|---|
| Enabled | Toggle the agent on or off |
| Tone | Professional, Friendly, Clinical, or Empathetic |
| Verbosity | Concise, Balanced, or Detailed |
| Custom instructions | Add practice-specific instructions (up to 2,000 chars) |
| Allowed actions | Restrict which actions the agent can suggest |
| Blocked actions | Prevent specific actions from being suggested |
| Require approval for | Set actions that always require manual approval |
| Daily request limit | Cap the number of requests per day for this agent |
Practitioner-Level AI Settings
Individual practitioners can customise their AI experience:
- Go to Settings > AI Preferences
- Configure options such as:
- Live note generation during sessions (on/off)
- Clinical guidance suggestions (on/off)
- Default note template for AI drafting
- Preferred language style for AI-generated content
These settings cascade: platform defaults apply first, then organisation overrides, then practitioner preferences.
Approval Requirements
All AI-generated suggestions require human approval before they take effect. This is a non-negotiable safety requirement aligned with PsyBA professional standards (2025).
- AI agents can draft, suggest, and recommend -- but never execute clinical or administrative actions autonomously
- Approval requests appear in the practitioner's task list
- The approval decision and reasoning are recorded in the audit trail
sole practice owner, owner-tier actions are auto-allowed without requiring a separate approver. :::
Data Processing Consent
Before AI features can process client data:
- The practice must enable AI processing in Settings > AI
- Individual clients must have AI processing consent recorded (collected via the consent management module)
- Voice transcription features require separate recording consent
- Couples therapy requires a consent model selection (no secrets, secrets, or case-by-case)
Conversation Retention
Configure how long Praxi keeps conversations that are not linked to a client:
- Navigate to Settings > AI
- Under Data & permissions, choose Conversation retention: 7 days, 30 days, 90 days, 180 days or 1 year (default 90 days)
- Conversations not linked to a client are deleted after they have been idle for longer than the period you choose
Conversations linked to a client are kept with that client's record (7 years, or until the client turns 25) and are not deleted by this setting. They can be released if the client asks for their information.
Choosing a shorter period than the current one asks you to confirm by typing SHORTEN, because conversations idle for longer than the new period will be deleted sooner and cannot be recovered.
AI Audit Trail
All AI interactions are logged separately from conversation history:
- Which agent was invoked and which model was used
- The input provided (with PHI redaction in logs)
- The output generated
- Whether the output was approved, modified, or rejected
- Who reviewed and actioned the AI suggestion
Audit logs are retained for 7 years in accordance with AHPRA requirements, regardless of the conversation retention setting.
View the AI audit trail at Settings > AI > Activity Log or in the main audit logs under the AI category.
Permissions
| Action | Required Permission |
|---|---|
| Configure AI settings | Practice owner |
| Enable/disable agents | Practice owner |
| Set practitioner preferences | Each practitioner (own settings) |
| View AI audit trail | Practice owner or supervisor |