AI configuration is where you decide how your agent thinks. There are four big knobs and a handful of smaller ones — this guide covers each in plain language.
Find these settings under Settings → AI configuration on any agent.
Model
The model is the underlying brain. Chatzuri supports the latest frontier models from multiple providers — GPT-class, Claude-class, Gemini-class, and open-source options.
- Frontier models (top-tier): smartest, best at nuance, more expensive per message. Use these for sales, complex support, and customer-facing flows where quality matters most.
- Mid-tier models: 80% of the quality at a fraction of the cost. Great default for most agents.
- Fast / cheap models: ideal for simple FAQ bots, high-volume routing, or background tasks.
Temperature
Temperature controls how creative or predictable the agent's answers are. The slider runs from 0 to 1.
- 0.0 – 0.3 — deterministic. Same question, same answer. Best for support, FAQ, anything factual.
- 0.4 – 0.7 — balanced. Slight variation in phrasing. Best for sales, conversation, light creativity.
- 0.8 – 1.0 — creative. Each answer phrased differently. Best for brainstorming, copywriting, and demo agents.
System prompt
The system prompt is the most important setting in the entire app. It's the instruction your agent reads before every conversation. It defines:
- Who the agent is and what it does
- The tone and voice it uses
- The boundaries — what it won't talk about
- How it should escalate when it doesn't know
A solid starter template
You are [Agent Name], the [role] for [Company]. You help customers with [scope].
Tone: friendly, concise, and helpful. Never use jargon. Never invent facts.
If a customer asks something you don't know, say so honestly and offer to escalate to a human via the support form.
Always end your response with a useful follow-up question when appropriate.Language
Pick the primary language your agent should respond in. If a customer writes in a different language, modern frontier models will usually switch automatically — but the Language setting nudges the agent's default.
Advanced knobs
- Max tokens — caps how long an answer can be. Useful for SMS where messages are charged per segment.
- Top-p — alternate to temperature for tuning answer diversity. Most teams leave it at default.
- Frequency / presence penalty — discourage word repetition. Useful for copywriting agents.
- Streaming — show answers as they're generated. On by default; turn off only for SMS-style channels.
- Function calling — required for agent tools and custom actions to work. Leave on unless you have a reason to disable.