Parameters Reference
Guide to AI model configuration parameters.
Current Implementation
The SF Explorer AI integration currently uses default parameters for all AI calls. Model selection is the primary way to control output behavior.
Supported Parameters
| Parameter | Status | Notes |
|---|---|---|
| Model Selection | ✅ Supported | Choose model via Settings or API |
| Prompt | ✅ Supported | Your input text |
| Temperature | 🔜 Planned | Not yet configurable |
| Max Tokens | 🔜 Planned | Not yet configurable |
| Top P | 🔜 Planned | Not yet configurable |
Model Selection (Primary Control)
Since advanced parameters aren't yet configurable, choose the right model for your use case:
| Use Case | Recommended Model | Why |
|---|---|---|
| Quick answers | GPT-4o Mini, Claude 3 Haiku | Fast, cost-effective |
| Complex reasoning | GPT-4.1, Claude Sonnet 4 | Better accuracy |
| Code generation | GPT-4o, GPT-4.1 | Optimized for code |
| Creative content | GPT-5, Claude Sonnet 4.5 | Higher creativity |
Parameter Concepts (Reference)
Understanding these concepts helps when parameters become configurable:
Temperature (0.0 - 2.0)
Controls randomness/creativity:
0.0 ━━━━━━━━━━━ 0.7 ━━━━━━━━━━━ 2.0
Deterministic Balanced Creative
- 0.0 - 0.3: Facts, data extraction
- 0.4 - 0.9: General use
- 1.0 - 2.0: Creative writing
Max Tokens
Limits response length:
- 1 token ≈ 4 characters ≈ 0.75 words
- 100 tokens ≈ 75 words
| Setting | Tokens | Use Case |
|---|---|---|
| Short | 100-300 | Quick answers |
| Medium | 300-800 | Standard responses |
| Long | 800-2000 | Detailed explanations |
Top P (Nucleus Sampling)
Controls token selection diversity (0.0 - 1.0):
- 1.0: Consider all tokens (default)
- 0.9: Top 90% probability mass
- 0.5: More focused responses
Related Documentation
- Model Comparison - Choose the right model
- Temperature Guide - Deep dive into temperature
- Best Practices - Testing tips
Last Updated: January 2026