AI models and parameters

Projects and agents use LLM configuration to control model behavior.

Common parameters

temperature

• Low (e.g. 0–0.3): More consistent, deterministic answers — support, policy text.
• Mid (e.g. 0.5–0.8): Balanced creativity.
• High (e.g. 0.9+): More varied phrasing — risk: more hallucinations.

maxTokens

• Caps response length and cost; very low values can truncate answers.

topP

• Nucleus sampling; works with temperature to control diversity.

Choosing a model

• Consider latency, cost, multilingual support, and tool-calling (function calling) compatibility.
• The exact model list depends on your backend / provider configuration.

Related

Projects · Agents