Documentation

AI assistant

Assistant identity and scope

Set your assistant's name, brand, tone and rules, understand what scope means, and learn why the AI stays silent until an identity with a brand is saved.

Last updated: 2026-09-29

The assistant identity is who your customers think they are talking to: a name, a brand, a tone and a few rules. Sela will not answer customers until you have set one up, so this is the first AI page to complete.

Why a saved identity is required#

Sela refuses to answer under its own name. On every message it checks that an active assistant profile exists and that the profile has a brand name. If either is missing the AI stays silent and your team sees the message in the inbox as usual.

Where to configure it#

Open Automation → Catalog & AI → AI identity in the dashboard sidebar. It needs the AI Assist add-on. Without it the page shows a locked state with an Upgrade button. See Plans and billing.

SectionFieldLimitWhat it does
BasicsAI name100 charactersThe name the assistant uses when it introduces itself.
BasicsBrand100 charactersThe business the assistant speaks for.
ToneTone presetFriendly, Professional, Warm, DirectThe overall voice.
ToneResponse styleConcise, Balanced, DetailedHow much it writes.
PersonalityPersona summary300 charactersA one-line description of the character.
RulesAlways do300 characters per ruleShort instructions the assistant must follow.
RulesNever do300 characters per ruleThings it must avoid.
Extra GuidanceAdditional instructions4,000 charactersFree-text guidance on style and procedure.

Live Preview shows a sample greeting in English and Arabic. It is a fixed sample of how your name, brand and tone combine, not a real AI reply, so always test with a real message too.

Steps#

  1. Open AI identity.
  2. Under Basics, enter the AI name and your brand.
  3. Choose a tone preset and response style.
  4. Add a persona summary if you want a specific character.
  5. Add rules as short imperatives, for example "Reply in Iraqi Arabic when the customer writes Arabic" or "Never use emojis".
  6. Press Save.
  7. Send a test message in the web widget or a test WhatsApp number and read the reply.

You can keep several profiles, duplicate one, choose which is the default, or archive the ones you do not use. The default profile cannot be archived. The AI uses the default active profile.

Other places to edit the same text#

Settings → Inbox settings → AI Assist edits only the additional instructions ("AI Assist personality") and holds the workspace-wide Disable AI replies switch. AI prompts lists your profiles with their additional instructions. All of them edit the same field, so the last save wins.

The "Stay grounded in trusted knowledge" switch on AI Assist is always on and cannot be turned off.

What scope means#

Scope is the set of things the assistant will talk about: your business only. Sela has a built-in rule that the assistant helps only with your brand's business, and it declines unrelated requests such as programming help, investment advice or homework, briefly and politely.

Scope is a boundary, not a suggestion:

  • Knowledge base text, campaign messages, earlier assistant replies and available tools cannot widen it.
  • Persona text and extra instructions cannot widen it either.
  • If you want the assistant to cover a new topic, give it real information for that topic, in the knowledge base or catalog.

What you can and cannot configure#

You canYou cannot
Name, brand, tone, style, persona, rules, extra guidanceChange the underlying AI
Run several profiles and pick the defaultMake the AI answer under Sela's name
Switch AI replies off for the whole workspaceGive the AI permissions or tools through persona text
Switch WhatsApp AI automation on or offChange platform-level limits or the automatic handoff policy
Turn on automatic quality reviewTurn off the "grounded in trusted knowledge" behavior

Two related settings do not currently affect replies, so do not rely on them: the "Configurations" block on the AI identity page (flow instructions and quick suggestions) and the "Programmed messages" editor on the Environment page. The Be Me persona, available only to workspaces the Sela team enables, is also stored but does not change what the AI says. Put anything you need enforced in Rules and Extra Guidance instead.

Writing good rules#

  • Put style here and facts elsewhere. Tone, formality, dialect, sign-offs and emoji rules belong in the identity. Prices, hours, policies and phone numbers belong in the knowledge base, catalog and business records.
  • Keep rules short and testable. "Answer in two sentences unless asked for detail" works. "Be amazing" does nothing.
  • Do not mention prices or promotions. Stale numbers in instructions conflict with live catalog data.
  • Do not try to override platform rules ("always give a discount", "never hand off"). They are ignored, and a reply that breaks them is not sent.
  • Do not repeat the same rule in three places. Contradictions lower quality.

Example of a compact rule set for a restaurant:

  • Always do: greet returning customers by name if known. Confirm the delivery area before quoting a fee.
  • Never do: mention competitors. Promise a delivery time.
  • Extra guidance: Keep replies under three sentences. Use Iraqi Arabic with Arabic customers and English otherwise. Sign off with the restaurant name only on the first message.

Pausing the AI#

To stop the AI everywhere, turn on Disable AI replies in Settings → Inbox settings → AI Assist. This also cancels replies already being generated. Humans can still reply. Turning it back on does not send the replies that were stopped. To stop it in a single conversation, use Pause assistant in the conversation header. See How the AI works.

Automatic quality review#

The AI identity page has an Automatic quality review card, off by default. It scores a percentage of completed replies each day, each review costing credits. Results appear in Reports under AI quality. See Improving answer quality.

Assistant identity and scope | Sela docs | Sela