Configure Chat Behavior Settings

Overview

Flow ID: configure-chat-settings
Category: Settings & Configuration
Estimated Duration: 3-5 minutes
User Role: All Users
Complexity: Moderate

Purpose: Adjust global chat behavior settings that affect how the AI responds, including temperature (creativity), frequency penalty (repetition reduction), lookback size (conversation memory), and dataset query parameters.

Trigger

What initiates this flow:

• User manually initiates

Specific trigger: User wants to customize AI behavior, typically because:

• Default settings don’t produce desired results
• Want more creative or more factual responses
• Need to adjust conversation memory
• Optimizing for specific use cases

Prerequisites

Before starting, users must have:

• Application running
• Basic understanding of chat settings impact

User Intent Analysis

Primary Intent

Customize global AI behavior settings to achieve desired response characteristics across all new conversations.

Secondary Intents

• Optimize response quality
• Control creativity vs. factuality
• Manage context window usage
• Configure dataset query behavior

Step-by-Step Flow

Main Path (Happy Path)

Step 1: Navigate to Settings

• User Action: Click “Settings” in navigation
• System Response: Settings page loads

Step 2: Access Chat Options Tab

• User Action: Click “Chat Options” tab
• System Response: Chat configuration options display
• UI Elements Visible:
  • Model selection (chat model, embedding model)
  • Context window size slider
  • Lookback size slider
  • Temperature slider (0-100)
  • Frequency penalty slider (-200 to 200)
  • Dataset query settings (number of results)
  • Save button (if not auto-save)

Step 3: Adjust Temperature

• User Action: Move temperature slider to desired value
• System Response:
  • Slider moves
  • Current value displays
  • Description may update (e.g., “Higher values = more creative”)
• UI Elements Visible:
  • Temperature slider
  • Value display (e.g., “70”)
  • Range: 0 (deterministic) to 100 (very creative)
  • Guidance text explaining effect
• Visual Cues: Slider position indicates current value

Step 4: Adjust Frequency Penalty

• User Action: Move frequency penalty slider
• System Response: Value updates
• UI Elements Visible:
  • Slider (-200 to 200 range)
  • Current value
  • Explanation: “Reduces repetition in responses”
• Visual Cues: Slider with neutral (0) in middle

Step 5: Set Lookback Size

• User Action: Adjust lookback slider (conversation memory)
• System Response: Value changes
• UI Elements Visible:
  • Slider (1-50 messages range)
  • Current value (e.g., “14 messages”)
  • Explanation: “How many previous messages AI remembers”

Step 6: Configure Dataset Query Settings

• User Action: Adjust number of dataset results slider
• System Response: Value updates
• UI Elements Visible:
  • Slider (1-10 results range)
  • Current value (e.g., “5 results”)
  • Explanation: “Number of document passages to include”

Step 7: Save Settings

• User Action: Click “Save” if required (or auto-saves)
• System Response:
  • Settings saved to database
  • Success notification may appear
  • Settings active immediately for new chats
• UI Elements Visible:
  • Success message
  • Save button may show “Saved” state briefly

Final Step: Settings Configured

• Success Indicator:
  • All settings at desired values
  • Settings persist after refresh
  • New chats use these settings
• System State Change:
  • Global chat settings updated
  • Applies to all new conversations
  • Existing conversations unchanged (unless modified individually)
• Next Possible Actions:
  • Test settings in new chat
  • Adjust further if needed
  • Configure per-chat settings for exceptions

Alternative Paths & Strategies

Strategy A: Reset to Defaults

When to use: Settings became problematic, want factory defaults

Steps:

1. Navigate to Chat Options
2. Click “Reset to Defaults” button (if available)
3. All settings revert to recommended values
4. Save if required

Strategy B: Per-Chat Custom Settings

When to use: Want different settings for specific conversation

Steps:

1. Set global defaults in Settings
2. Open specific chat
3. Access advanced settings in chat
4. Override specific settings for this chat only
5. Chat uses custom settings; others use global

Error States & Recovery

QA Note: Settings are constrained to valid ranges. Invalid values prevented by UI controls (sliders). Technical errors unlikely.

Pain Points & Friction

Identified Issues:

1. Settings Require Understanding AI Concepts

   • Impact: Non-technical users uncertain what values to use
   • Potential Improvement:
     • Presets: “Creative”, “Balanced”, “Precise”
     • Recommendations based on use case
     • Examples of setting effects

2. No Preview of Settings Impact

   • Impact: Must test in real conversation
   • Potential Improvement:
     • Live preview with sample response
     • Show example outputs for different values
     • A/B comparison

3. Global Settings Override Complex

   • Impact: Per-chat overrides not obvious
   • Potential Improvement:
     • Clearer hierarchy explanation
     • Visual indication when using custom

Design Considerations

Following Contextual Design Principles:

1. Automation Opportunities: Recommend settings based on usage patterns
2. Simplification Opportunities: Preset configurations
3. User Trust: Clear explanations of each setting
4. Cognitive Load: Reduce technical terminology

Related Flows

• Configure Context Window Settings - Related setting
• Entourage Mode Manage Chat Personas - Per-chat overrides
• Create New Empty Chat - Uses these settings

Technical References

Knowledge Base Sections:

• src/components/chat-options-tab/index.js - Settings interface
• src/components/chat/chat-advanced-settings.js - Per-chat overrides
• src/localdb/settings.js - Settings persistence

Version History

Notes

Settings Explained:

• Temperature: 0 = predictable/factual, 100 = creative/varied
• Frequency Penalty: Negative values allow repetition, positive reduce it
• Lookback Size: How many recent messages AI considers (more = better context, slower)
• Dataset Results: More results = more context but slower processing

Best Practices:

• Start with defaults (temperature: 70, frequency penalty: 0, lookback: 14)
• Adjust one setting at a time to understand impact
• Use higher temperature for creative tasks, lower for factual
• Increase lookback for complex conversations needing lots of context
• Use 3-5 dataset results for balance of speed and coverage

Common User Questions:

• “What should I set temperature to?” - 50-70 for balanced, 80-100 for creative, 0-30 for factual
• “Why do settings have numbers instead of labels?” - Provides fine-grained control; presets would simplify
• “Do settings affect existing chats?” - No, only new chats unless overridden per-chat
• “What’s safe to experiment with?” - All settings; can always reset to defaults