Skip to main content
One assistant reaches people through five channels: a website widget for anonymous visitors on your own pages, an HTTP API for your own product or bot, an MCP server for Claude, IDE assistants and agent frameworks, a Slack bot (beta) that answers in the thread, and the console for your own team. Every channel reaches the same assistant, the same knowledge and the same grounding, so an answer is the same whichever channel it arrives through, and an abstention on one is an abstention on all. What differs is the credential and who is asking.

​
Where to set each one up

In the console, everything under Deploy is about reaching your assistant.

​
Choosing a channel

Use the widget. It is one script tag, works for visitors who never sign in, and is bound to the origins you list.
Use the HTTP API from your backend. Your server holds the API key and relays answers and citations to your interface. Stream them for a responsive experience.
Use the MCP server. Claude, IDE assistants and other MCP clients can search your knowledge and ask your assistant, with citations.

​
Usage by channel

The usage report breaks questions down by channel: user for the console, service for API keys (including MCP clients) and widget for embeds.

​
Not available

There are no webhooks, no SDK and no command-line tool. Webhooks and polling, Client libraries and Command line show what to use instead. Chat, Agents and the Model Gateway are separate surfaces of the product, not channels for an assistant: see Other surfaces.
Last modified on November 11, 2026