Profile
Profile is the first sub-page under AI Agent → Configuration → General, and it's where you set who your bot is and how it greets people. Because Configuration is project-level, everything you set here applies to all your agents - both conversational agents and guided agents. Get the profile right and every reply, from any agent, inherits the same persona and identity.
This page covers three things, in the order we recommend filling them out:
- Pick a persona - set the tone with a preset.
- Write the bot identity - the system instruction that defines who the bot is.
- Set the welcome message - how the first turn of every conversation begins.
To open it, go to AI Agent → Configuration in your bot and click Profile settings in the left menu.
Step 1: Pick a persona
Choose one of the available presets - for example Polite and persuasive, Empathetic and helpful, or Witty. The persona is a quick way to set tone without writing prompt text yourself.
Best practices
- Pick one persona and let your bot identity (next step) carry the nuance. Mixing presets in your head ("I want it warm but witty") usually leads to a less coherent voice.
- Match persona to channel. Voice flows benefit from Empathetic and helpful; sales pages often want Polite and persuasive.
Write the bot identity
This is the most important field on the page. Bot identity is the system instruction that tells your bot who it is, what it does, and what it must not do.
Write it as if you were briefing a new hire on day one:
You are Aria, the support assistant for Acme Logistics. You help customers
track shipments, file claims, and update delivery preferences. You never
quote pricing - that's handled by the sales team. If a customer asks about
pricing, route them to the @sales-agent. Use a calm, professional tone.
Do not make up tracking numbers under any circumstance.
Best practices
- Lead with the role, then the responsibilities, then the constraints, then the tone. The model pays the most attention to your first lines.
- Be concrete. "You help customers" is too vague; "You help customers track shipments and file claims" is testable.
- State what the bot must NOT do, not just what it should do. The "never quote pricing" instruction in the example above is what stops bad behavior.
- Don't write marketing copy here. This is an instruction to a model, not a brand brochure.
Tip -
@-mentions. Type@in this field (or any prompt field) to insert a live reference to an agent, workflow, or tool. The chip stays linked to the target's slug, so it doesn't break when the target is renamed. Mentioning is a hint to routing - it does not trigger a handoff. To make an agent actually take over, use a Routing Logic rule.
Step 3: Set the welcome message
Still on Profile settings, scroll to the Choose how to welcome section. Click Change to open the picker - Nexus shows three kinds of options:
| Option | Studio label | When to use |
|---|---|---|
| Instruction | Instruct super agent | Generate a greeting from a prompt fragment you write. Good for personalised welcomes (returning user vs new, channel-specific tone). |
| Send message | Send message | A fixed greeting with optional quick-reply chips. Best for simple welcomes. |
| A welcome workflow | (workflow name, e.g. runWelcomeWorkflow) | Pick any workflow registered for welcome use. Use this when the greeting needs to do something first (look up a customer, show rich media). The picker lists each workflow as its own option. |
The default for fresh bots is Instruct super agent. Quick-reply chips ("Conversation starters") are available in both Instruct and Send message modes - they appear under the message body on the same Profile settings page.
The two non-Instruct options skip the LLM. Both Send message and welcome workflows are deterministic - they emit a pre-built or workflow-rendered message without a model round-trip on the welcome turn. Only Instruct super agent generates the welcome via the LLM. Pick a workflow option when you need to do something on welcome (look up the customer, render rich media) without the cost or latency of an LLM call.
If you choose a static response, you can add quick reply chips so users have suggested next actions.
Best practices
- Keep welcomes short. Long welcomes scare users away on chat.
- Show 2-4 quick-reply chips for the most common journeys. More than 4 becomes noise.
- If your welcome needs personalisation but no lookups, Instruction is usually the right choice. For lookups (loyalty status, account info), do them in a workflow tool called after the welcome.
Next: Conversation Rules - the persona-level always do / never do guidelines applied to every reply.