Copilot Overview
A Copilot is an embeddable AI chat widget you can add to any website. It’s one of the three app types in Apps, alongside Web and Mobile. Instead of generating an application, a Copilot puts one of your existing agents behind a branded chat window and hosts it for you at a public URL.
A Copilot is powered by an agent you’ve already built. Before you start, you’ll want a published AI Agent or Multi-Agent System to act as its “brain.” See Build an AI Agent.
How a Copilot differs from Web and Mobile apps
Web and Mobile apps are generated — you describe them in a prompt and qRaptor writes the code, provisions a database, and builds the app. A Copilot works differently:
| Web / Mobile app | Copilot | |
|---|---|---|
| How you build it | Free-text prompt → generation pipeline | Short guided wizard |
| Database | Provisioned (Postgres) | None |
| Generated code | Yes — full-stack app | None — it’s configuration |
| What answers users | Your app’s own logic | A bound AI Agent or Multi-Agent System |
| Where it runs | A workspace pod | qRaptor’s hosted Copilot renderer |
| How you ship it | Publish → deploy the built app | Deploy → embed a snippet on your site |
Because there’s no code to generate or build, creating a Copilot is fast, and configuration changes go live within about a minute — no rebuild.
The Copilot lifecycle
Every Copilot follows the same path, and this guide is organized around it:
- Create — Name it, pick its brain, set a starting brand color.
- Studio tour — Get oriented in the Copilot Studio: five tabs and a live preview.
- Bind a brain — Connect an AI Agent or a Multi-Agent System — the thing that answers users.
- Appearance — Colors, logo, header text, welcome message, and custom CSS.
- Behavior — Suggested prompts, threads, and message limits.
- Test — Chat with it in the live preview and inspect the event stream.
- Deploy — Pick a URL and who can use it, then go live.
- Embed on your site — Drop in a script or iframe and manage allowed origins.
What powers a deployed Copilot
When a user chats with your deployed Copilot, messages flow from the widget to the bound agent and back:
- The chat UI is served by qRaptor’s centralized Copilot renderer — it reads your Copilot’s configuration (brand, behavior, brain) and renders the widget.
- Each message is routed to the bound AI Agent or Multi-Agent System, which runs with its tools, memory, and reasoning exactly as it does everywhere else in qRaptor.
- Human-in-the-loop steps, streamed responses, and tool activity all surface in the chat, just as they do in the agent preview.
A Copilot is a thin front end over an agent. If you want to change what it knows or can do, edit the underlying agent — not the Copilot. The Copilot controls how the chat looks and where it’s embedded.