Train your AI

Open Automations in your dashboard, create one (or open the default), and you'll see three configuration areas. Customers often go live with just the defaults; the three knobs below take it from "decent" to "obviously yours."

1. Personality

The personality block shapes how the AI talks. Five fields do most of the work:

• Role — one sentence describing what the AI is. "Friendly support agent at a running-gear store." "Booking concierge for Acme Salon." The more specific, the better.
• Tone — friendly / professional / formal / playful. Pick one or describe in your own words.
• Length — short replies (1–2 lines), medium (3–4 lines), or detailed. Most channels feel best on short.
• Emoji usage — none, occasional, or expressive. Match your brand voice; Instagram tolerates more than B2B WhatsApp.
• Language — English, Turkish, or "match the customer." The AI mirrors customer language automatically when set to match-the-customer.

2. Knowledge base

The knowledge base is where you put static facts about your business — things that don't live in Shopify or Google Calendar but customers ask about constantly. Typically 10–30 entries is enough.

What goes in:

• Return / refund policy (and any exceptions)
• Shipping times and methods
• Business hours and contact info
• Locations (with addresses and parking notes)
• Common pricing FAQs (especially deal-breaker questions like minimum orders, deposits, discounts)
• Brand-voice rules ("we never call customers 'guys'", "we don't make medical recommendations")
• Anything you find yourself answering 5+ times a week that isn't in your product catalog

Each entry is a Q/A pair or a free-form note. The AI consults these on every message, alongside personality and linked RAG sources.

3. Linked data sources (RAG)

For anything live — product catalog, inventory, calendar availability, CRM contacts — link a RAG source instead of copying data into the knowledge base. See the full list at all data sources.

Inside the automation editor, the Linked data sources section shows every RAG source connected to your workspace. Check the boxes for the sources this automation should use. Most workspaces link one or two — over-linking actually hurts response quality because the AI burns time on irrelevant queries.

Test your AI before going live

Every automation has a built-in Test chat in the editor — type questions as if you were a customer and see how it replies, using the same configuration that'll run in production. Tweak personality and knowledge base until the test replies feel right.

When you save, the new configuration applies to every active channel immediately. No restart needed; new incoming messages use the new config.

When to split into multiple automations

One automation per workspace handles most cases. Split when:

• You serve two distinct customer types. Example: B2C support and B2B sales need different personalities, different knowledge bases.
• You run multiple languages with different brand voices. Same AI, two automations — one EN, one TR.
• You run multiple brands from one workspace (agencies on the Agency plan).

Each automation can be assigned to specific channels, so you can route by source.