Developer docs: install, configure and integrate Shukai

Shukai is an assistant that sits in the corner of a website or an application, answers from the owner’s own content, opens pages, fills in forms and presses buttons, talks out loud, and hands people over to a person. The owner sets it up in the panel without code. What a developer adds is the one line that installs it and, where wanted, a little markup and a few event listeners.

Three levels, in the order you need them

Level 1 — install. One <script> tag, pasted from the panel’s Install tab. That is the whole installation for a website: the widget reads the page’s language, answers from the site the owner indexed, and performs navigation itself. Install the widget.

Level 2 — make it yours. Theme tokens on the host element, your own launcher button, a few lines of JavaScript to open the widget or ask it something, listeners for the events it emits when it acts. The widget.

Level 3 — let it act and let it call you. Mark up the forms it may fill and the button that saves; declare the screens of your app; describe one HTTP call to your API so it answers from live data; name an event your page renders. Page actions, Integrations.

Two APIs

• The runtime is what the widget speaks to: a WebSocket for the conversation, a few HTTP endpoints for speech, recognition and beacons. It is authenticated with the site’s public widget key and restricted by origin. You need it only for a custom front end. Runtime API.
• The management API is what the panel speaks to: sites, sources, agents, widgets, tools, conversations, analytics. It is authenticated with a secret management key. You need it to automate the panel. Management API.

Ask the assistant

Every page of this documentation is in the assistant’s own knowledge base. The widget in the corner answers questions about it — «how do I mark the button that saves?» — and opens the page that says so.

What is where