Your AI agent
Your agent has one brain across every channel. Its behavior is controlled by its instructions — a system prompt written in plain language — plus the knowledge OneChat gathered from your website and connected platforms.
How the agent is built
- Your website. When you paste your URL at signup, OneChat reads your site, FAQs and policies and drafts the first version of the instructions for you.
- Your platforms. Connected channels enrich the knowledge — a Shopify store contributes products, policies and order data.
- Your instructions. Everything you write under Agent → Instructions takes priority. This is where you set tone, add policies, and draw hard lines.
Writing good instructions
Treat the instructions like onboarding notes for a new support hire:
- Tone: "Friendly and concise. Use the customer's name. No emoji."
- Policies the website doesn't state: "Orders placed after 2pm JST ship the next business day."
- Hard lines: "Never promise a delivery date. Never offer a discount code. If asked, offer to check with the team."
- Escalation triggers: "If the customer mentions a refund over $200 or sounds upset, hand off to a human."
tip
Iterate with real questions. Send your agent the ten questions your team answers most, read its replies in the dashboard, then tighten the instructions where it drifted. Two or three rounds of this is usually enough.
What the agent does on its own
- Detects language and replies natively in 55+ languages — see Languages.
- Transcribes voice notes on LINE and WhatsApp — see Voice notes.
- Knows when to stop. When it's unsure, or the customer asks for a person, it escalates to your team on Slack instead of guessing.
Updating the agent
Instructions changes take effect immediately on every channel — edit once, updated everywhere. There is no per-channel configuration to keep in sync.