Skip to main content
Besides LINE, Facebook, and Instagram, a customer agent can answer inside your own app—an inline chat popup, a help panel, a full support page. Every thread shows up in the Conversation Pool as a web conversation, right next to your other channels, with the same draft review, takeover, and handoff controls.
Unlike LINE and Facebook, the web channel needs no pairing under Channels. Any app in your org can talk to any of the org’s agents—your app names the agent when it sends.

​
How it works

  1. Create a customer agent in the dashboard (Creating an agent). That’s the only dashboard setup.
  2. Your app sends the customer’s message, naming the agent and a thread key you manage (one per chat session or per user).
  3. Your app polls for the agent’s reply and renders it. A typing indicator tells you when the agent is composing.
The agent’s reply mode still applies: in auto-draft mode the reply is staged for your team to approve before it appears in the thread.

​
Building it

The chat surface lives in your app, backed by the pre-installed customer SDK (@stardeck-customer-apps/integrations-sdk, server-side only—credentials are auto-injected). There’s no prebuilt widget component; your app owns the UI, so you can make it a floating popup, a sidebar, or a full page. The SDK handles sending messages, polling for replies (with a real typing indicator), image attachments, and—if you want a support console inside your own app—the full operator surface for draft approval and human replies.
The easiest way to build this is to ask the developer agent to “add a support chat widget backed by the assistants SDK.” It knows the SDK and the send-then-poll pattern and will wire up the routes and UI for you.

​
Limits & notes

  • Rate limit — 5,000 messages per day per deployment.
  • No pairing, no per-app scoping — any deployed app in the org can reach any of the org’s agents.
  • Same pool, same controls — web conversations support the same draft review, takeover, and handoff flow as every other channel (Operator workflow).