Product support that answers from your docs and escalates with context

The agent answers only from what you gave it. Every reply starts with a search of your workspace's knowledge base: the docs pages you imported, uploaded files, Q&A pairs, synced Notion or Drive pages and published Help Center articles. It does not browse the web and it does not answer from general knowledge about your product. When nothing relevant comes back, it says so and offers to bring in your team rather than guessing at an API parameter.

Docs sites suit the crawler. It keeps a page's main content and drops the navigation, sidebar, header, footer and cookie banner, so menu text never competes with the actual answer. A page behind a login or rendered only by JavaScript is seen the way an anonymous first request sees it; upload that content as a document instead.

Agent instructions shape the voice without touching the facts. 'Keep answers short and point to the relevant doc' goes in How to sound and behave; 'Never promise a release date' goes in Never do this. The base rules that keep the agent honest sit above both, so an instruction cannot make it invent an answer or skip the handoff.

A bug report in the chat becomes a Jira issue with the visitor's email attached. With Jira connected, the agent writes a clear summary and the details, files a Task in the project you chose, and tells the visitor the issue key. There is deliberately no action for customers to check an issue's status: your tracker is for your team.

Slack and Discord take the urgent cases. When a visitor reports an outage, a failed payment or a request from a key account, the agent posts 'Support alert from' the visitor's name with a one or two sentence summary to the channel you picked, after it has escalated or resolved what it can. Discord alerts have mentions disabled, so nobody can make the bot ping everyone.

If your team lives in Zendesk, Freshdesk, Zoho Desk, HubSpot or Salesforce, the agent can create a ticket or case with a subject and the details, and in all of those but HubSpot check an existing one by number. Every action card has a Test button that runs the real request, and actions you add with a key start switched off until you flip them on.

Escalation is quiet in the widget and loud for your team. When a visitor asks for a person, accepts the agent's offer of help, or something fails, the conversation turns escalated, the agent stops replying, and every workspace member gets an email with the visitor's name and email when known, the last three messages and an Open conversation button. At most one email per conversation every six hours, so a visitor who asks twice does not page the team twice.

In the inbox, the full transcript is there before anyone types. Assign the conversation to a teammate from its header, leave an internal note the visitor never sees, and use Enhance to tidy a draft before sending. Your reply shows up live in the widget under your display name.

Set Team availability to business hours and the agent stops promising a quick human reply outside them. It gives your away message with the day and time the team is back, and asks for an email address so you can follow up. The widget's status line flips between team online and team offline on its own.