Profile settings
Profile settings sets who your assistant is and how it opens a conversation. It's the first section under AI agents → Configuration.
Everything here is inherited by every agent you build, single and multi alike. The panel says the division plainly - "Inherited by every agent. What each agent does is set by its own Goal." Profile settings decides how your assistant sounds; each agent's Goal decides what it does.
Identity & tone
Four fields. The first two are what customers see; the last two are what the model reads.
Name and Company
| Field | What it's for |
|---|---|
| Name | Your agent's name across voice, chat and WhatsApp |
| Company | The business your agent represents |
These two are customer-facing, so they carry brand weight. They also feed the Insert example templates below, so fill them in first.
Identity
One or two sentences on who the assistant is and who it represents. The placeholder shows the register:
You are Aria, a support assistant for Acme Insurance.
It must be between 20 and 600 characters. Short enough that it can't become a second instructions field, long enough to be more than a name.
Not sure how to start? Insert example offers three templates, filled in with the Name and Company you just set:
| Template | Shape |
|---|---|
| Customer support | You are {name}, a customer support assistant for {company}. |
| Brand assistant | You are {name}, a virtual assistant representing {company}. |
| Employee helpdesk | You are {name}, {company}'s internal employee helpdesk assistant. |
What belongs here: who the assistant is, which company it represents, and the one-line scope of what it covers.
What doesn't: step-by-step behaviour, tool instructions, or policy. Those belong in an agent's own instructions, where they apply to the step that needs them. Anything here is sent on every turn of every agent, so it has to be true for all of them.
Good: "You are Aria, the support assistant for Acme Logistics. You help customers with shipments, claims and delivery preferences."
Too vague: "You are a helpful assistant." Adds nothing the tone preset didn't.
Wrong field: "You are Aria… Always check the order status tool before answering, and if the customer is angry, offer a refund." That's procedure - it belongs in the agent that does it.
Tone
How the assistant sounds. Two presets and a custom option:
| What it sends to the model | ||
|---|---|---|
| Helpful (default) | Warm & clear | "Be warm and empathetic, acknowledging the user's concerns when they come up. Keep your answers clear and easy to understand." |
| Professional | Formal & precise | "Be calm and professional. Keep your wording clear and natural." |
| Write your own | Your text, sent as-is |
The presets are exactly those two lines - nothing hidden. If neither fits, Write your own and describe the voice in a sentence or two; the box is seeded from whichever preset you were on, so you can start from one and adjust. It can't be left blank.
Tone applies to every reply from every agent. Set it once here rather than repeating tone instructions in each agent's prompt, where copies drift apart.
Channels
A read-only summary of the channels this assistant is live on, with Manage channels to configure them. Nothing is set here - it's a pointer to Channels & Deploy.
Choose how to welcome
How the first turn of every conversation begins. Three options, in the order they appear:
| Option | What happens | Model call? |
|---|---|---|
| Prompt the agent | You write a short instruction; the assistant generates a greeting from it | Yes |
| Run welcome flow | A flow delivers a rich, multi-step welcome | No |
| Send message | Fixed wording you type, sent verbatim | No |
The distinction that matters: only Prompt the agent involves the model. The other two are fixed - the same words every time, no latency, no cost, no chance of drift. Choose one of those when the opening must be exact (a compliance line, a fixed menu), and Prompt the agent when it should adapt.
Reach for a welcome flow when the greeting needs to do something first - look up the customer, check an account state, render a card - without an LLM round-trip.
Conversation starters are quick-reply buttons shown under the welcome. They're available with Prompt the agent and Send message, but not with a welcome flow. Two to four covering your most common journeys is the useful range; more reads as a menu.
Keep welcomes short. A long opening message is the most common reason people abandon a chat before saying anything.
What lives elsewhere
Some things you might expect on this page are their own Configuration sections:
- Conversation rules - the shared always-do / never-do guidelines applied to every reply.
- Memory - the variables your agents can read and write.
- Fallback - what happens when the assistant can't answer.
- AI safety & conduct - content filters and guardrails.
Next: Conversation rules - the guidelines every reply follows.