Overview

Sigma Domain uses multiple AI providers and models, each optimized for different tasks. The platform intelligently selects the best model for your request, or you can choose manually.

Available AI Providers

Anthropic Claude

• Strengths: Complex reasoning, long-form analysis, coding, careful instruction following
• Models: Claude Opus (most capable), Claude Sonnet (balanced), Claude Haiku (fastest)
• Best for: Analysis, coding, document editing, complex questions

OpenAI GPT

• Strengths: General intelligence, creative writing, function calling
• Models: GPT-4o (flagship), GPT-4o-mini (faster/cheaper)
• Best for: Creative content, general tasks, structured outputs

Google Gemini

• Strengths: Multimodal understanding, vision, long context
• Models: Gemini Pro, Gemini Flash
• Best for: Image analysis, multi-modal tasks, large document processing

Model Selection

Auto Mode (Recommended)

• Sigma Domain automatically selects the best model for each request
• Different parts of your request may use different models
• The tool track, image track, web track each use optimized models
• This is the default and recommended setting

Manual Model Selection

1. Click the model selector in the top bar
2. Choose a specific model from the dropdown
3. All subsequent messages will use that model
4. Switch back to "Auto" to restore automatic selection

Model Roles

The system uses different model "roles" internally:

• Primary - Main conversation model (typically Claude or GPT-4)
• Fast - Quick classification and simple tasks (Haiku or GPT-4o-mini)
• Vision - Image analysis tasks (GPT-4 Vision or Gemini)
• Code - Code generation and analysis

How Model Routing Works

When you send a message:

1. Classification - A fast model determines the type of request
2. Track Selection - The multi-track system activates relevant tracks
3. Model Assignment - Each track uses its optimal model:
   • Support detection: Fast model
   • Web search synthesis: Primary model
   • Image analysis: Vision model
   • Function parameter inference: Fast model
   • Final response: Primary model

Per-Feature Model Usage

Chat Responses

• Uses the primary model (or your manual selection)
• Auto mode typically uses Claude for reasoning, GPT for creative tasks

Tool Execution

• Parameter inference uses a fast model
• Tool selection uses the primary model
• Error analysis may use a specialized model

Document Editing

• Document analysis uses the primary model
• Edit generation uses the primary model
• Format-specific operations may use specialized models

Jobs/Workflows

• Planning uses the primary model
• Step execution uses appropriate models per step
• Claude Code sessions (for support) use Haiku by default

Cost and Speed Considerations

Faster Models

• Haiku, GPT-4o-mini, Gemini Flash
• Lower cost per token
• Faster response times (under 1 second)
• Good for simple queries, classification, summaries

More Capable Models

• Opus, GPT-4o, Gemini Pro
• Higher cost per token
• Slower but more thorough responses
• Better for complex analysis, coding, reasoning

Auto Mode Optimization

• Balances speed and quality automatically
• Uses fast models for simple tasks
• Upgrades to capable models for complex requests
• Minimizes cost while maintaining quality

Troubleshooting

Response Quality Too Low

• Switch from a fast model to a more capable one
• Use Auto mode to let the system choose
• Provide more context in your question

Response Too Slow

• Switch to a faster model (Haiku, GPT-4o-mini)
• Shorter prompts process faster
• Check your internet connection

Model Not Available

• Some models may be temporarily unavailable
• Auto mode will fall back to an available model
• Check the model selector for current availability

Wrong Model Being Used

• Check if you have a model manually selected
• Switch to Auto mode to restore automatic selection
• Some features always use specific models regardless of selection