You already shipped an assistant in your product. Then, because support still had to live somewhere, you pasted a chat widget on the same page. Now, there are two places to ask for help. The customer picks whichever is closer to the cursor, and the transcript stays in the one you aren't looking at.
That's when I'd stop adding widgets.
Headless customer support means they never leave your UI. The assistant, the help article, the "can I talk to someone" thread, the contact form. Those are your components, and they live wherever you want them to. Fernand is the inbox behind them: we handle the routing, the contact, the person who answers.
Is a widget still fine?
I wouldn't build a fully headless support process on day one. If your product doesn't have an assistant yet, a support widget on the marketing site and on the page where people get stuck is more than enough. I already wrote that process up in how to set up support as a founder.
The widget starts to feel wrong when the product already talks. If you have any kind of agentic capabilities in your product, then this is where having an extra layer support widget just doesn't make sense for the customer experience. Your customers don't care which org pieces of the product belong to, they just want to report a bug, request a feature, or clarify a billing issue. Don't get in the way.
The handoff
Your assistant answers first. When it's stuck, or the customer just says they want a person, it opens a conversation in Fernand with the transcript.
What the customer should not get is a "we emailed you" and a fresh thread to go hunt for. Russ asks about a double charge. Jared joins the same chat, already knowing what was said, and refunds it from the inbox. That's the feature. The rest is plumbing, and plumbing is allowed to be boring.
Fields, signature, and a prompt you can hand to Claude Code or Cursor are on headless customer support. I'd let the agent write the first version, and make sure to instruct it to plug it into what you have built, and use your own design components (e.g. Tailwind, Shadcn..)
What else can you build headless?
Once that handoff works, the same inbox can sit behind more of the product. I wouldn't do the rest in the same week. You'll still be debugging the first one, and that's a better use of the afternoon.
A headless support portal for someone who's signed in, where they see their own conversations and reply. Your layout.
Embedded in-context help articles you write in Fernand and render yourself, on the two or three screens where people actually get stuck. Not a documentation project. Three screens.
A headless support tab in your admin. Whoever is already staring at the account can read the thread without opening another tool. This is the one I wish we'd had sooner. The context switch is the job, most days.
Fernand is free to start. The API comes with it. We don't charge per call, per resolution or when the assistant hands someone to you. Try it out and see how you like it!