Branding controls how the widget looks. Persona controls how it sounds. The two are separate pages because they are separate jobs, and the Persona page keeps the same live preview so you can see the agent name, avatar and opening message as you write them.
Wording only, never the facts
Everything on this page shapes how an answer is phrased. None of it changes what the assistant is allowed to claim, where it gets its data, or the rules it will not break. A persona cannot talk the assistant into giving financial advice or into stating something it did not read.
Tone
Pick a starting voice. This sets the default register: how long answers run, how formal they are, how much hand-holding they include.
Custom tone of voice
Optional, and it layers on top of the tone above. Describe how your assistant should talk in your own words: personality, warmth, phrases to use or avoid, British or American spelling, emoji or none. Be specific. "Warm and a little playful, British spelling, never corporate, no exclamation marks" produces a far more consistent voice than "friendly".
Agent name and avatar
The name shown as the widget's title, and used for the avatar initials when no image is set. Leave it blank to use your project name. The avatar takes any square image URL.
Opening message
The first thing a user sees. Leave it blank for the default greeting. Note that when a user connects a wallet, the assistant may lead with what it can already see rather than this greeting: if their last transaction failed, it says so instead of waiting to be asked.
Disclaimer
Shown under the chat box and attached to every escalation, so it reaches whoever reads the ticket later. Unlike every other field here, blank does not mean "use the default": clearing it removes the disclaimer entirely. Leave it as it is unless your own legal wording says otherwise.
Clearing the disclaimer removes it
The assistant still refuses financial, tax and legal advice, which is not switchable by any protocol, but nothing on screen will say so.
Suggested questions
Starter questions shown as tappable chips in the chat. Set your own and they replace the assistant's automatic suggestions, so users are only ever offered questions you already know it can answer. This is the cheapest way to steer people towards what you have documented and away from what you have not.
Write them as a user would
"Why did my transaction fail?" beats "Transaction troubleshooting". The chips become the user's first message, so they should read like something a person would actually type.
Beta programme
Set up on its own Beta programme page in the dashboard. For running the assistant with testers rather than customers. It keeps everything it already does, including live transaction diagnosis, and adds what testers need on top. It is off by default and changes nothing until you turn it on.
Introduces itself
The panel opens once per visit and leads with your introduction, so a tester who has just landed is told what to try instead of facing an empty page. Desktop only, and never if they have already opened it themselves.
Feedback
A button, not a guess. For opinions, suggestions and reactions. The assistant stops trying to solve, asks only what the tester expected to happen, and records it with the conversation attached. It never promises anyone will reply, because you cannot answer every tester individually.
Bug reports
A separate button for things that are broken, switched on independently of feedback. The assistant asks what they were trying to do and where, then stops: it never asks for their browser, wallet, network or a transaction hash, because the report already carries all of it.
Introduction
What testers see when it opens itself. Leave it blank to use your normal opening message. The tappable starter questions come from Suggested questions above.
End date
Optional and recommended. After it passes the assistant goes back to normal on its own, so a welcome to the beta never greets someone months later.
Findings land where your team works
Bugs and feedback appear on the Findings page in the dashboard, each with the conversation that produced it, and route into Slack, Linear, GitHub or Jira like any other escalation, with the page, wallet and chain attached. They are kept out of the support ticket queue, because nobody leaving a note is waiting for a reply.