Personalization
Clatri already knows who you are — from your data, your conversations, and your entity description. What's left is to decide how you want it to speak to you.
The agent personality defines its tone, response style, and communication rules. It's a free-text field in the settings, separate from the entity description. If the description answers who you are, personality answers how you want to be spoken to.
Professional and concise:
Be direct and concise. Use a professional but friendly tone.
Give me practical examples when explaining something.
Minimalist:
Short answers. No emojis. No fluff.
Analytical:
When talking about finances, always include percentages and comparisons.
Use tables when it makes sense. Prioritize data over opinions.
Personality is per-entity. You can have an informal tone for your personal entity and a professional one for your company. If you don't configure any, the agent uses its default behavior.
Preferences you want to keep
You can also ask in chat for Clatri to keep a preference, such as “Remember that I prefer brief answers” or “My health reports should stay only in chat.” These preferences are stored separately from conversation summaries. When you correct a preference, the new instruction replaces its previous value.
If you ask to add or correct a personality instruction, the agent preserves the others; say explicitly if you want to replace the entire personality. A confirmation should correspond to a saved change, rather than just a reply saying “I'll remember.”
Channel restrictions are also considered when delivering Schedule results. An email may contain only a chat link when the content should stay there.
Clatri can proactively update your personal description and behavioral defaults when you share clear, lasting information that improves its assistance. It preserves compatible instructions, tells you what it saved, and lets you review or correct it in Settings. A one-off action does not automatically become a general rule. These fields are included automatically as context; they do not depend on memory search.