Enable, configure, and set defaults for your domain
Default AI Model
—
Supporting AI Models
—
Total AI Model Catalog
—
Loading models...
No models found matching your criteria.
Effective System Instructions
ℹ️This shows the complete system instructions that this particular AI model will use for this tenant.
Model Configuration
ℹ️This shows the complete system instructions that this particular AI model will use for this tenant.
⚠️This model uses an OpenAI Assistant. Instructions are configured in the OpenAI Assistants dashboard. Custom domain instructions cannot override assistant settings.
Optional friendly name that describes this model's specialty or role in your domain (e.g., "Legal Expert", "Creative Writer")
Optional description of what this model is used for in your domain
Sales-oriented description shown in onboarding modals when users select this model. Explain what makes this model special and when to use it.
Custom instructions for this model in your domain. Define how the AI should behave and respond.
This is THE system prompt used for this model in your domain.
ℹ️
Per-tenant parameter overrides for your default model.
Leave fields blank to use provider defaults (NULL in database). Settings are scoped to this domain only.
📘 Parameter Guide & Best Practices
🌡️ Temperature (0.0–2.0)
Controls randomness in word choice:
• Lower (0.0–0.3): Deterministic, terse, formal. Good for policy text, compliance, code.
• Medium (0.4–0.8): Balanced, natural tone. General-purpose Q&A.
• Higher (0.9–1.3): Creative, varied, chatty. Brainstorming.
• Very high (1.4–2.0): Highly diverse; expect occasional off-track output. 💡 For professional assistants: 0.2–0.5 reads professional without stiffness.
📏 Max Tokens (positive integer)
Hard cap on output length (tokens = words + word pieces):
• Short replies/alerts: 256–512
• Detailed answers: 800–1200
• Long summaries/drafts: 1500–2000+ ⚠️ Smaller = faster & cheaper, but risks truncation. Larger = fuller answers, more cost.
🎯 Top P (0.0–1.0)
"Nucleus sampling" - samples from smallest set of words with cumulative probability ≥ p:
• Lower (0.1–0.5): Conservative wording (like lower temperature).
• Higher (0.8–1.0): Diverse wording (like higher temperature). ⚠️ Important: Use Temperature OR Top-P, not both. They overlap in effect. When in doubt, leave Top-P blank and tune Temperature only.
💰 Cost & Performance: Grows with output length (Max Tokens) and sometimes with higher creativity. 🔒 Determinism: Set Temperature = 0.0 for most repeatable outputs (slight variance may still occur). ✂️ Truncation: If answers cut off mid-sentence, raise Max Tokens or ask for shorter answers.
🧠Upload and manage Knowledge Nodes specific to this model. Knowledge Nodes are procedural, modular bundles that augment the RAG system with executable logic.