Home AI Agent

AI Agent

The AI assistant and Copilot
Super Admin
By Super Admin
7 articles

What the AI Agent is

What the AI Agent is The AI Agent is Kulpunai's built-in AI assistant. It answers customers on its own using your documents and FAQs, and hands the conversation to an agent when it cannot help. You will find AI Agent in the sidebar. What it is made of Assistants A configured AI personality: instructions, tone, welcome message and handoff message. You can have several — say, one for sales and one for support. FAQs Question-and-answer pairs. The most precise source of knowledge: when a customer's question matches an FAQ, the assistant answers exactly as you wrote it. Documents Files, links and spreadsheets the assistant learns from: product descriptions, delivery terms, policies. Scenarios Step-by-step instructions for recurring situations, such as processing a return. Tools Let the assistant reach into your systems — check an order status, look a customer up in your CRM. Playground A test chat where you check the assistant's answers without exposing it to real customers. Inboxes The channels the assistant works in. Until an inbox is connected, customers get no AI replies. Settings Instructions, guardrails and response guidelines — the boundaries the assistant works within. Copilot A separate mode for agents: it summarises a thread, drafts a reply, and looks up contact and conversation data. Customers never see Copilot. What you need to run it The AI Agent is available on paid plans. Usage is measured in credits: one AI reply or one voice message transcription costs one credit. Your balance is shown under Settings → Billing. Getting it running 1. Create an assistant — see "Creating an assistant". 2. Upload documents about your product. 3. Check its answers in the Playground. 4. Connect the assistant to an inbox. 5. Keep reviewing the FAQs it suggests from real conversations.

Last updated on August 23, 2026

Creating an assistant

Creating an assistant Quick setup The simplest route is Quick setup under AI Agent → Assistants. The platform asks four questions and writes the assistant's instructions for you: 1. Company or product name. 2. What the company does — for example, "we sell handmade textiles with nationwide delivery". 3. What the assistant should help with — "answer questions about products, prices, delivery and returns". 4. Tone of voice — friendly, professional or concise. Click to generate the instructions, review the result and edit it if needed — this is the assistant's system prompt. Then create the assistant. Main settings Assistant settings live under AI Agent → Settings. Basic information - Name — how the assistant introduces itself; - Description — what it is for; - Product name. System messages - Welcome message — the first message in a conversation; - Handoff message — what the assistant says when passing the conversation to a human; - Resolution message — the text used when a conversation is closed automatically. Instructions The main text that shapes the assistant's behaviour. Describe what it should do, what it must not do, and how to behave in edge cases. Response temperature The lower the value, the more focused and predictable the answers. Support usually calls for a low temperature. Features Toggles that extend what the assistant does: - Generate FAQs from resolved conversations — the assistant proposes new question-and-answer pairs based on real threads; - Store key details as memories — the assistant remembers what matters about a customer between conversations; - Extract contact attributes from conversations — it fills the contact card from the conversation; - Include source citations in replies — the assistant points to the document an answer came from. Guardrails and response guidelines Two limiters are available in the assistant settings: - Guardrails define the topics the assistant must not answer. Anything outside them is handed to an agent. - Response guidelines define structure and style: brief or detailed, formal or friendly. Auto-resolve The assistant can close conversations where the customer stopped replying. Turn on automatic resolution of inactive conversations and pick a period from 15 minutes to 24 hours. Deleting Deleting an assistant removes it from every connected inbox and permanently deletes its data.

Last updated on August 23, 2026

Playground and connecting to channels

Playground and connecting to channels Playground AI Agent → Playground is a test chat with the assistant, where you check its answers without exposing it to real customers. Worth testing before launch: - the questions customers actually ask — prices, timings, availability; - off-topic questions, where the assistant should politely decline and hand over; - phrasings where the customer makes a typo or gives incomplete information; - questions whose answer only exists in a document, to confirm the assistant finds it. Playground messages consume AI Agent credits just like replies to customers. Connecting to channels 1. Open AI Agent → Inboxes. 2. Tick the inboxes where the assistant should reply. 3. Save. The assistant can also be attached from the inbox settings themselves. How the assistant and agents work together The assistant replies first. When a question falls outside its knowledge, or a guardrail is triggered, it sends the handoff message and passes the conversation to an agent. An agent can step in at any point simply by replying in the conversation. Launch recommendations - Start with one channel. The website chat widget or Telegram are the easiest. - For the first week, read through conversations the AI handled and fix the documents wherever it got things wrong. - Make sure agents are online during business hours: a handoff is no help if nobody picks it up. - Watch your credit balance so the assistant does not stop mid-day.

Last updated on August 23, 2026