AI Configuration
Creativity Level (Temperature)
The Creativity Level slider controls how predictable or inventive the AI's replies are. Think of temperature as a creativity dial — lower values give more predictable, factual responses, while higher values produce more creative and varied replies. Technically, this maps to the AI model's temperature parameter.
| Setting | Temperature | Behavior |
|---|---|---|
| Low (Conservative) | 0.1 – 0.3 | Very predictable, sticks closely to facts and templates. Best for legal, medical, or financial replies where accuracy is critical. |
| Medium (Balanced) | 0.4 – 0.6 | Good mix of accuracy and natural-sounding language. Recommended default for most businesses. |
| High (Creative) | 0.7 – 1.0 | More varied and creative phrasing. Good for marketing or casual conversations, but may occasionally produce unexpected wording. |
Recommendation: Start at Medium (0.5). Only increase if replies feel too robotic, or decrease if they feel too unpredictable.
Confidence Level
The Confidence Level setting defines the minimum confidence score an AI reply must achieve before it is considered ready. This directly affects auto-send behavior:
- High (85–100): Only very confident replies are auto-sent. More replies go to manual review. Safest option.
- Medium (65–84): Moderate threshold. A balanced approach for established knowledge bases.
- Low (50–64): More replies are auto-sent, but with higher risk of inaccurate responses. Only use if you have thoroughly tested your Knowledge Base.
Navigate to Settings → AI Configuration → Confidence Level to adjust. The confidence score for each reply is visible in the conversation thread next to the AI badge.
Reply Length Settings
Control how long or short AI-generated replies should be:
- Concise: 1–3 sentences. Quick, direct answers. Ideal for support tickets with simple questions.
- Standard: 3–6 sentences. Provides context and explanation. Recommended default.
- Detailed: 6–12 sentences. Thorough responses with step-by-step instructions. Good for technical support.
- Custom: Set a specific maximum word count or token limit.
The AI treats this as a guideline, not a hard limit. A "Concise" reply to a complex question may still be a few sentences longer than usual if the answer genuinely requires it.
Language Settings
Configure how the AI handles language in replies:
- Auto-detect (default): The AI detects the language of the incoming email and replies in the same language.
- Fixed language: Force all AI replies to use a specific language (e.g., English, Spanish, German), regardless of the incoming email language.
- Primary + fallback: Reply in the detected language if supported, otherwise fall back to your primary language.
Configure at Settings → AI Configuration → Language.
Greeting and Sign-off Customization
You can define custom greetings and sign-offs that the AI will use in every reply:
- Greeting: e.g., "Hi {first_name}," or "Hello," or "Dear {first_name},"
- Sign-off: e.g., "Best regards,\n{agent_name}\n{company_name}" or "Thanks,\nThe {company_name} Team"
Available placeholders:
| Placeholder | Replaced With |
|---|---|
{first_name} | Recipient's first name (from contact record or email) |
{last_name} | Recipient's last name |
{full_name} | Recipient's full name |
{agent_name} | Name of the assigned agent or account owner |
{company_name} | Your company name from settings |
Max AI Replies Per Conversation
To prevent infinite AI reply loops (especially in auto-send mode), you can set a maximum number of AI replies allowed in a single conversation thread. Once the limit is reached, new emails in that thread are flagged for human review.
- Default: 5 replies per conversation.
- Range: 1 to 50 (or unlimited).
This is especially important when two automated systems are emailing each other. Without a cap, they could exchange hundreds of messages in minutes.
AI Spending Cap Setup
AI API calls cost money. MailTrixy lets you set a spending cap so you never get a surprise bill from your AI provider.
- Go to Settings → AI Configuration → Spending Limits.
- Set a daily limit (e.g., $5/day) and/or a monthly limit (e.g., $100/month).
- Choose what happens when the limit is reached:
- Pause AI replies: Stop generating AI replies until the next period.
- Switch to cheaper model: Automatically downgrade from GPT-4o to GPT-4o-mini.
- Notify only: Send an alert but continue generating replies.
MailTrixy tracks token usage for every AI request and displays real-time cost estimates on the dashboard.
Monthly Cost Limit Configuration
The monthly cost limit is a hard cap on total AI spending across all users on your MailTrixy instance (for SaaS operators) or for your account (for single-tenant users).
- Admin-level limit: The system administrator sets a global monthly budget. Once reached, AI features are paused for all users until the next billing cycle.
- Per-user limit: In SaaS mode, each subscription plan can have its own AI budget. For example, the Basic plan might allow $10/month in AI usage while the Pro plan allows $50/month.
- Usage dashboard: Both admins and users can see a real-time breakdown of AI costs by provider, model, and day.
Configure at Admin Panel → Plans → Edit Plan → AI Limits for per-plan limits, or Settings → AI Configuration → Global Limits for system-wide limits.