Giving the agent a personality

The Agent page, under Aura AI, is where you decide how the agent thinks: the manner it answers in, the job it is doing, and what it knows about you. It holds exactly seven settings, and it is worth saying plainly what it does not hold, because that is the question people arrive with.

The name, greeting, languages, colours, logo, avatar and support contacts are not here. They belong to the widget, and they are set under Channels → Web chat. There is one editor for those, not two. There are also no model or LLM settings on this page at all — the model is not something a workspace chooses.

Editing needs admin or owner. Everybody else sees the same three sections as a read-only list, with "Not set" where a value is empty.

How it answers

This card is the register, not the appearance.

• Reply tone — a choice from Conversational, Professional, Friendly, Formal, Empathetic and Playful. This is the register every reply is written in. If your agent was set up elsewhere with a tone outside that list, Auralis keeps it rather than quietly coercing it to one of the six.
• Persona — free text. The character the agent plays: its manner, not its job. "Warm, unhurried, never cute" is a persona.
• Brand voice — free text, up to 1000 characters. Phrasing dos and don'ts the agent follows. This is the right place for "never say unfortunately", "we call them members, not users", "no exclamation marks".

Persona and brand voice sound similar and pull in different directions in practice. Persona is who is speaking. Brand voice is the house style they are speaking in. If you only fill in one, fill in brand voice: it produces the more consistent change.

What it is for

• Agent role — its job title, up to five words, for example "customer support specialist". The counter enforces the limit as you type.
• Agent goals — up to 250 characters describing what a good answer achieves. Not a list of tasks; a definition of done.

What it knows about you

• Company info — what your company does, in the agent's own words. This is background the agent answers from over and above its training data, so it is the place for facts too small or too general to be worth an article.
• Keyword glossary — a list, built one entry at a time with Add. Product names and terms the agent should spell exactly this way. This is the setting that stops an agent writing "Auralis AI" when the product is "Aura AI", and it earns its keep faster than anything else on the page.

Saving

The form saves as a whole. The button stays inactive until something has actually changed, and it stays inactive while any field is over its ceiling — Auralis will not send a partial save behind your back. If you are over a limit it names the field and says so: nothing is sent while a field is over its ceiling.

A successful save says "Saved. These are the agent's live values now." A failed one keeps everything you typed. Nothing is cleared, so you adjust the named field and save again rather than retyping the form.

Only the fields you changed are sent. That matters if two admins are editing at once: the last write wins per field rather than per form, so you will not silently overwrite a colleague's untouched field with a stale copy of it. The page records the time it read the values, near the save bar, so you can tell how old what you are looking at is.

When the page will not open

If Auralis cannot be reached, the page shows a card saying so rather than an empty form. That is deliberate. A form rendered without live values would happily save blanks over the agent's real configuration, and a page that cannot read is a page that must not write.