Tutorial Customizing Mosaic

Tutorial: Customizing Mosaic

This tutorial shows you how to tailor Mosaic to your specific needs — from writing effective system prompts and tuning temperature to choosing themes and optimizing your provider setup.

Estimated time: 15 minutes

---

Step 1: Write an Effective System Prompt

The system prompt sets the AI's behavior, tone, and constraints. It's the single most powerful customization.

Accessing the System Prompt

1. Open Settings (⚙️)
2. Find the System Prompt textarea

Example System Prompts

For technical documentation:

You are a senior technical writer. Write clear, concise documentation
with code examples. Use active voice. Include a "Prerequisites" section
for any setup steps. Always explain WHY before HOW.

For creative brainstorming:

You are a creative brainstorming partner. Generate diverse, unconventional
ideas. Don't self-censor. Structure responses as: (1) The core idea,
(2) Why it might work, (3) Potential challenges, (4) How to get started.

For code review:

You are a senior software engineer conducting a code review. Analyze code for:
- Correctness: does it work?
- Performance: can it be optimized?
- Maintainability: is it readable and well-structured?
- Security: are there vulnerabilities?
Be constructive and specific in your feedback.

Tips for Good System Prompts

Do	Don't
Be specific about tone and format	Use vague instructions like "be helpful"
Include constraints (length, format)	Assume the AI knows your context
Set clear role expectations	Overwhelm with contradictory instructions
Keep it under 500 characters	Write a novel — concise prompts work better

---

Step 2: Tune Temperature

Temperature controls randomness in the AI's responses:

Value	Range	Effect	Best For
0.0 - 0.2	Low	Deterministic, focused, repetitive	Code generation, factual Q&A, data extraction
0.3 - 0.7	Medium	Balanced creativity and precision	General conversation, analysis, explanations
0.8 - 1.2	High	Creative, diverse, unpredictable	Brainstorming, creative writing, ideation
1.3 - 2.0	Very High	Highly random, may be incoherent	Experimental, poetry, surreal content

Setting Temperature

1. Open Settings (⚙️)
2. Adjust the Temperature slider (0.0 to 2.0)

Practical Tips

• Code generation: Use 0.1-0.2 for deterministic, correct code
• Debugging/analysis: Use 0.3-0.5 for clear, logical analysis
• Explanations: Use 0.5-0.7 for natural, readable explanations
• Brainstorming: Use 0.8-1.0 for diverse ideas
• Switch mid-conversation: You can change temperature between branches for different perspectives on the same topic

---

Step 3: Choose a Theme

Mosaic has 5 visual themes. The right theme reduces eye strain and improves focus.

Theme Gallery

Theme	Type	Background	Accent	Best For
Void	Dark	Deep black/blue	Cool blue	Long coding sessions, low-light environments
Dusk	Dark	Cosmic purple	Rich purple	Creative work, evening use
Sand	Light	Warm beige	Earth tones	Daytime use, bright rooms
Snow	Light	Clean cool gray	Blue-gray	Professional settings, readability
Sunrise	Light	Golden warm	Gold tones	Morning use, warm environments

Switching Themes

1. Open Settings (⚙️)
2. Click a theme in the Theme grid
3. The theme applies immediately

Theme Auto-Detection

On first launch, Mosaic auto-detects your OS theme preference (dark/light) and selects the appropriate theme. You can override this manually.

Accessibility

Mosaic respects your OS accessibility settings:

• prefers-reduced-motion: Disables animations
• prefers-reduced-transparency: Reduces glass effects

---

Step 4: Optimize Provider Configuration

Provider Selection Strategy

Goal	Recommended Provider	Model
Maximum quality	Anthropic	Claude Sonnet 4
Best value	Mistral	Mistral Large
Speed/simplicity	Mistral	Mistral Small
Creative tasks	OpenAI	GPT-4o
Local/offline	Ollama	Any local model
Cost-sensitive	Ollama or Mistral Small	Local or Small

Multiple Provider Setup

Configure all providers for maximum flexibility:

1. Mistral: Primary — fast, reliable, good value
2. OpenAI: Secondary — good for creative tasks
3. Anthropic: For complex reasoning tasks
4. Gemini: Alternative for specific tasks
5. Ollama: Offline/local when internet is unavailable

Debate Model Configuration

For parallel debate, choose models with complementary strengths:

Combination	Best For
Mistral Large + GPT-4o	General comparison
Claude Sonnet + Mistral Large	High-quality reasoning
Mistral Small + GPT-4o-mini	Cost-effective comparison
All 4 providers	Comprehensive analysis

---

Step 5: Configure Canvas Preferences

Minimap

Toggle in Settings → Canvas → Minimap:

• On (default): Shows an overview of all nodes in the bottom-right corner
• Off: More canvas space, less visual distraction

Confidence Scoring

Toggle in Settings → Canvas → Confidence:

• On (default): Each response shows a confidence badge
• Off: Cleaner node appearance, saves one API call per response

Suggestion Tendrils

Toggle in Settings → Canvas → Tendrils:

• On (default): Auto-generated follow-up suggestions after each response
• Off: No suggestion nodes, less visual activity

---

Step 6: Master Auto-Layout

Auto-layout arranges nodes in a radial tree pattern. Useful when:

• The canvas is messy after extensive branching
• You've dragged nodes around and want a clean arrangement
• After importing a canvas from another session

Trigger auto-layout:

1. Settings → Auto-arrange (arranges entire canvas)
2. Right-click a node → Auto-arrange (arranges from that node downward)

---

Step 7: Export/Import for Portability

Export Workflow

1. Complete your analysis
2. Settings → Export
3. File downloads as mosaic-{canvas-name}.json
4. Share with team members or archive

Import Workflow

1. Settings → Import
2. Select a .json file
3. Canvas is validated and loaded
4. All nodes, edges, bookmarks, and analytics are restored

Backup Strategy

• Daily: Export critical canvases at the end of each session
• Before major changes: Export before trying experimental features
• Sharing: Export a clean canvas without sensitive data

---

Step 8: Customize with Ollama Local Models

For full control, run local models via Ollama:

1. Install Ollama
2. Pull models: ollama pull llama3.2 ollama pull mistral
3. Point Mosaic to http://localhost:11434
4. Click Detect

Benefits of local models:

• No API costs — runs entirely on your machine
• Offline — works without internet
• Privacy — data never leaves your computer
• Custom models — use any model Ollama supports

Tradeoffs:

• Slower than cloud APIs (especially on CPU)
• Lower quality for most local models vs. cloud
• Requires significant RAM/VRAM

---

Step 9: Create Workflow Templates

Once you've configured Mosaic perfectly, save your setup:

1. Create a blank canvas with your ideal system prompt
2. Configure your preferred theme, temperature, and provider
3. Export this canvas as template-default.json
4. When starting a new project, Import this template

You can create multiple templates for different use cases:

• template-code-review.json: Low temperature, Mistral Large, system prompt for code
• template-brainstorm.json: High temperature, GPT-4o, creative system prompt
• template-research.json: Medium temperature, RAG enabled, analytical system prompt

---

Quick Reference: Settings Summary

Setting	Location	Range/Options	Default
API Keys	Settings → API Keys	Mistral, OpenAI, Anthropic, Gemini	Empty
Ollama URL	Settings → Ollama	URL	http://localhost:11434
System Prompt	Settings → System Prompt	Free text	Empty
Temperature	Settings → Temperature	0.0 - 2.0	0.7
Theme	Settings → Theme	Void, Dusk, Sand, Snow, Sunrise	Auto-detected
Minimap	Settings → Canvas	On/Off	On
Confidence	Settings → Canvas	On/Off	On
Tendrils	Settings → Canvas	On/Off	On
Debate Models	Settings → Debate Models	Multi-select	None

---

Next Steps

• Advanced AI Features — Deep dive into each advanced feature
• Contributing Guide — Help improve Mosaic
• Keyboard Shortcuts and UI Reference — Complete UI documentation