1. Documentation
  2. AI support
  3. Configuring

Creating an agent

Name, avatar, personality, model.


Identity

  • Name and avatar: the agent speaks under its own identity, not Bao's. Your members see an interlocutor consistent with your brand.
  • This identity requires the Discord Manage Webhooks permission. Without it, the agent speaks under the bot's identity. If your server installed Bao before this feature existed, re-invite it.

Personality

A free-form text where you describe who it is and how it speaks. That is also where you set its scope.

A good personality says:

  • who it talks to (customers, players, beginners);
  • what it talks about, and what it does not;
  • how it talks (formal or informal, length, tone);
  • what to do when it does not know.

Example: "You are the shop's assistant. You answer in English, politely, in three sentences at most. You talk about orders, deliveries and returns. For any billing question, you hand over to the staff."

Model

You do not pick a model by its technical name, but a tier:

TierFor whom
EssentialFAQ, rules, everyday support
EconomyEveryday support, community, e-commerce
Standard (recommended)Paying customers, multi-step procedures
HigherTechnical fields, diagnostics, edge cases
MaximumHigh stakes, where a mistake is expensive

Standard is the default, and not the cheapest: an agent that misunderstands on the first try does not invite a second one. You can always step back down later.

An Advanced settings screen lets you name a specific model if you have a preference. The list is closed.

Vision

Optional: the agent can look at the images sent to it. Very useful for software or hardware support ("here is my screenshot"). Three images at most are taken into account per message, so a single reply cannot cost a fortune.

Sandbox

The dashboard lets you test an agent before exposing it: you talk to it, you see its replies, its sources and its cost, without any member being involved.