Your AI doesn't just answer. It acts.

An agentic skill lets Zipchat's AI agent act in a tool your store uses, not just answer: look up an order, create a discount code, open a ticket, book a call, or check stock in a store. The agent decides when to use the skill, asks for any missing details, runs the steps, and answers with the real result. It is the same agent that answers in website chat, WhatsApp, Instagram, Messenger, and email, so each customer keeps one conversation.

Both. With Agentic Skills, Zipchat does not only answer from your knowledge base: it calls your tools, reads live data, and runs the actions you set up, then answers with the result the tool returns. You decide which skills it has, on which channels, and which actions wait for the shopper to confirm.

Four. Order lookup inside Shopify finds an order and replies with its status and shipping; it is read-only and never cancels, edits, or refunds an order. Create discounts inside Shopify creates a unique, single-use percentage code up to the maximum you set; it is off by default. Escalate conversation hands a conversation to your team. Skip conversation keeps the AI from answering messages you describe, such as automatic emails from no-reply senders. Order lookup is on when Shopify is connected; turn the others on when you need them.

Any tool with an API. A gallery of ready-made templates covers Judge.me (submit a review), Zendesk (escalate to a human), WooCommerce (order tracking and discount codes), Freshdesk (create a ticket), Qapla' (order tracking), Clean Size Charts (size guides), and Cal.com (book a call): install the template, add your credentials, and test it. For anything else, such as an order management system, a returns platform, a carrier, a loyalty program, or an inventory system, you add your own skill. A skill calls the tool's API, or a webhook URL the tool gives you as an HTTP endpoint.

Usually not. A template needs your credentials and a test. For your own skill, you describe it in plain words: when to use it, when not to, and what to ask the shopper. Claude Code or Sidekick can help you write it, and you add the API details and keys from your tool. Most skills go live in under 30 minutes.

Each skill says in plain words when to use it and when not to. In a conversation, the agent checks whether the request matches a skill, asks for anything missing (an order number, an email, a size, a store), runs the steps, and answers with the result the tool returns. A skill can chain steps: check first, then act.

The agent answers with the result the tool really returns, so it does not say an action worked when it did not. If the tool returns an error, or the request is outside what the skill allows, it tells the customer. If Escalate conversation is on, it hands the case to your team.

Yes. The agent only uses the skills you turn on, on the channels you choose for each one. A skill that changes data can ask the shopper to confirm first, and a skill that shows personal data can verify the customer first. With Draft Mode your team approves replies before they are sent. You see each action in the conversation.

You choose per skill: website chat, WhatsApp, Instagram, Messenger, and email. The website chat setting also covers AI Search and product questions. Turn a skill on only where the action makes sense; some templates run on fewer channels.

Yes, in Test Chat. Test Chat calls the real API, so a skill that changes data can have real effects: test write skills with test records or a test store, check the result in your tool, then turn the skill on for your live channels.

Credentials and keys are stored encrypted as skill variables and are never written into replies. Give each skill a key with only the permissions it needs, and have the agent verify the customer before it shows personal data such as an order or a points balance. Zipchat follows GDPR and the EU-US Data Privacy Framework.

Car parts fitment by year, make, and model, showing only the parts the catalog lists as compatible; stock for one SKU in one physical store; two-step returns, where a read-only skill checks eligibility, then a second creates the return and label once the shopper picks the item; subscriptions that pause, skip, or resume after the shopper confirms; loyalty points and rewards after the customer is verified; accessory compatibility from an official product map; a bike size guide; draft orders with assisted payment; and pet appointments and vaccination records for verified customers. Each one is a custom skill a merchant set up, not a built-in feature.

A rules-based chatbot follows a fixed decision tree. It cannot read natural language or respond to questions it was not explicitly programmed to handle. Zipchat's agentic AI understands natural language, determines intent, and runs the right skill however the customer phrases the request.

A standard integration is a fixed pipe between two systems. It sends data from A to B when a specific event fires. An agentic skill is context-aware. The AI determines when to use the integration, what data to send, and how to communicate the result to the customer, all based on the live conversation.

They fit when your team answers the same order or account questions all day, when customers need live data rather than static FAQ answers, and when your tools have an API Zipchat can call. They are not the right fit if your tools have no API or other connection method, or if your compliance rules require a person to review every customer action.