Skip to main content

AI Configuration

ClaryNext's AI features power intelligent activity components, content generation, and personalized experiences. This guide covers AI configuration for administrators.

AI Overview​

AI Capabilities​

ClaryNext AI enables:

  • Personalized activity feedback
  • Content generation (Sketchbook)
  • Intelligent analysis
  • Dynamic recommendations
  • Conversational interactions

AI Components in Activities​

Authors can use:

  • AI Prompt (conversational)
  • AI Analysis (pattern recognition)
  • AI Content (dynamic generation)
  • AI Feedback (response evaluation)

Default AI Settings​

Accessing AI Configuration​

  1. Go to Admin > AI Configuration
  2. View and modify AI settings

Global Settings​

AI Enable/Disable

  • Turn AI features on/off platform-wide
  • Emergency kill switch
  • Maintenance mode

Default Model

  • Select default AI model
  • Configure fallback options
  • Set quality/speed tradeoffs

Rate Limits

  • Requests per user per hour
  • Daily limits
  • Burst allowances

Credit System

  • Credit cost per AI call
  • Free tier allowances
  • Credit refresh rules

Prompt Configuration​

System Prompts​

Configure default system behavior:

Base System Prompt

You are a supportive coach helping users with personal
development. Be encouraging, practical, and concise.
Keep responses under 200 words unless asked for more.

Context Variables Available variables in prompts:

  • {{user.name}} — User's name
  • {{activity.title}} — Current activity
  • {{organization.name}} — User's organization
  • {{response.X}} — Previous activity responses

Prompt Templates​

Create reusable templates:

  1. Go to AI > Prompt Templates
  2. Click New Template
  3. Define template:
    • Name
    • Description
    • System prompt
    • User prompt template
    • Output format
  4. Save for author use

Example Templates​

Coaching Feedback

Name: Coaching Feedback
Description: Provides supportive coaching feedback

System: |
  You are a supportive coach. Provide brief, encouraging
  feedback that acknowledges the user's insight and
  suggests one actionable next step.

User Template: |
  The user shared: "{{input}}"
  Context: They're working on {{goal}}.

Output: 3-4 sentences, encouraging tone

Reflection Prompt

Name: Deepen Reflection
Description: Helps users go deeper in reflection

System: |
  Ask one thoughtful question that helps the user
  explore their response more deeply. Be curious
  and non-judgmental.

User Template: |
  The user reflected: "{{input}}"

Output: Single question, open-ended

Knowledge Base​

Purpose​

The knowledge base provides:

  • Context for AI responses
  • Accurate information
  • Consistency across interactions
  • Domain-specific knowledge

Managing Knowledge​

  1. Go to AI > Knowledge Base
  2. View and edit knowledge entries

Adding Knowledge​

Manual Entry

  1. Click Add Entry
  2. Enter:
    • Topic/category
    • Content
    • Keywords
    • Priority
  3. Save

Import

  • Upload documents (PDF, MD, TXT)
  • Bulk import via CSV
  • API upload

Knowledge Structure​

Organize by:

  • Categories (topics)
  • Priority (importance)
  • Scope (global, org, activity)
  • Freshness (update frequency)

Example Entries​

Topic: Boundary Setting
Category: Communication
Priority: High
Content: |
  Boundaries are limits we set to protect our wellbeing.
  The DEAR MAN technique is effective for boundary
  conversations:
  - Describe the situation objectively
  - Express your feelings using "I" statements
  - Assert your needs clearly
  - Reinforce the benefits
  - Stay Mindful of your goal
  - Appear confident
  - Negotiate if needed

AI Examples​

Purpose​

Examples guide AI behavior:

  • Few-shot learning
  • Consistent output style
  • Quality benchmarks
  • Edge case handling

Creating Examples​

  1. Go to AI > Examples
  2. Click New Example
  3. Provide:
    • Input (what user said)
    • Ideal Output (desired response)
    • Context (when to use)
  4. Save

Example Entry​

Category: Goal Setting Feedback

Input: |
  I want to be more productive.

Bad Output: |
  That's great! You should try time blocking and
  setting priorities. [Too generic, not personalized]

Good Output: |
  That's a meaningful goal. To help you make it
  actionable: What does "productive" look like for
  you specifically? Is it about getting more done,
  feeling less stressed, or something else?

AI Guardrails​

Safety Settings​

Configure safety filters:

  • Content appropriateness
  • Sensitivity handling
  • Topic restrictions
  • Response limits

Content Policies​

Define what AI should not do:

  • Medical/legal advice
  • Personal predictions
  • Controversial opinions
  • Harmful suggestions

Fallback Responses​

When AI can't respond appropriately:

Default Fallback: |
  I appreciate you sharing that. This seems like
  something that might benefit from a conversation
  with a professional. Would you like to continue
  with the next step?

Monitoring​

Track AI behavior:

  • Response quality scores
  • Flagged responses
  • User feedback
  • Pattern detection

Organization-Specific AI​

Per-Organization Settings​

Organizations can have custom:

  • System prompts
  • Knowledge bases
  • Examples
  • Guardrails

Inheritance​

Settings cascade:

  1. Global defaults
  2. Organization overrides
  3. Activity-specific overrides

Enabling Customization​

Allow orgs to customize:

  1. Go to Organizations > [Org] > AI
  2. Enable customization
  3. Set bounds/limits
  4. Org admins can modify

AI Credits & Usage​

Credit Configuration​

Set credit costs:

  • Simple AI call: 1 credit
  • Complex analysis: 2-3 credits
  • Content generation: 3-5 credits

Allocations​

Define allocations:

  • Free tier: 10 credits/month
  • Basic plan: 50 credits/month
  • Premium: 200 credits/month
  • Unlimited options

Usage Monitoring​

Track usage:

  • By user
  • By organization
  • By activity
  • By feature type

Alerts​

Set usage alerts:

  • Approaching limit
  • Unusual patterns
  • Cost thresholds

Performance & Quality​

Quality Metrics​

Track AI quality:

  • User satisfaction ratings
  • Completion rates
  • Re-prompt rates
  • Feedback scores

A/B Testing​

If supported:

  • Test prompt variations
  • Compare model performance
  • Optimize over time

Continuous Improvement​

  • Review flagged responses
  • Update examples
  • Refine prompts
  • Expand knowledge base

Troubleshooting​

AI Not Responding​

Check:

  • AI enabled globally?
  • Credits available?
  • Rate limits hit?
  • Service status?

Poor Quality Responses​

Try:

  • Review/update prompts
  • Add relevant examples
  • Expand knowledge base
  • Adjust guardrails

Slow Responses​

Check:

  • Model selection
  • Response length limits
  • Network issues
  • Load status

Best Practices​

Prompt Engineering​

  • Be specific and clear
  • Provide good context
  • Include constraints
  • Test extensively

Knowledge Management​

  • Keep current
  • Verify accuracy
  • Organize well
  • Prune outdated content

Safety First​

  • Conservative guardrails
  • Regular audits
  • User feedback loops
  • Quick escalation paths

Next Steps​