cerver

Embeddable chat

cerver is the backend for your chat.

Add an AI chat to your site in two lines. Models, streaming, sessions, memory, multi-model routing — all handled. Don't worry about the backend. And that same backend runs many different agents: chat is one face of it.

Set up your chat cerver-chat on GitHub

✦ try it right here

This chat is running on cerver. So is everything on the page.

No sign-up to try — it's an anonymous trial session on cerver's backend. Pick a model, ask it something. This is the exact widget you'd embed in two lines.

01 — Two lines

Embed in two lines. The UI, the streaming, the store — all done.

Set the chat up in the dashboard, then drop these two lines where it should live. cerver renders the widget, streams the tokens, keeps the transcript, and routes the model. You build no backend.

<script src="https://cerver.ai/embed/cerver-chat.js"></script>
<cerver-chat pk="pk_live_…"></cerver-chat>

02 — Multi-tenant by default

Per-visitor sessions, scoped — your customers never see each other's threads.

The widget creates and remembers a session per visitor automatically. cerver scopes the transcript and the history to that session. No tenancy plumbing on your side.

Scoped transcripts

Every turn is durable and bound to one user. Read the tail, page by cursor, or download the full thread.

Resumable

Close the tab and come back — the session keeps the conversation so users pick up where they left off.

No PII required

The user id is a UUID; cerver never needs an email or a name to keep threads apart.

03 — Multi-model, live

Switch models mid-conversation. Claude, GPT, Gemini, Grok, local.

The widget has a model picker — switch the model without losing the thread. You're not locked to one provider; cerver routes each turn to whichever model you pick. Same session, same transcript underneath.

<!-- let visitors pick; or pin the list yourself -->
<cerver-chat pk="pk_live_…"
  models="Claude:claude-sonnet-4-6,GPT:gpt-5,Gemini:gemini-3.1-flash"></cerver-chat>
→ each reply is labeled with who answered; the thread stays bound

04 — One backend, many agents

The chat is one face. The same backend runs your other agents.

cerver-chat rides the same durable session layer cerver gives the CLI and the API. So the backend behind your chat is the backend behind coding agents, research fleets, and scheduled jobs — switch the chat's model live today, and grow into the rest without re-plumbing anything.

same backend runs
embeddable chat
coding agents
research fleets
scheduled jobs

05 — Designed by an agent

Describe the look. An agent writes the CSS.

No theming config, no design tokens to learn. Say "dark, rounded, like Linear" in the dashboard and a cerver agent restyles the widget for you. Change your mind later — describe it again; the same embed picks it up. No redeploy.

you: make it dark, rounded corners, our pink accent (#d2588f)
→ agent writes the CSS · saved to your chat · live on the next load

06 — Pricing

Want a chat product? Let's talk.

Pricing for the embeddable chat backend depends on your volume and how you embed it — so we size a plan around what you're building. Tokens always pass through at cost. Running cerver's control plane for your own team too? That's seat-based and separate.

Contact us for pricing → Set up your chat

§ it's live

It's live. Add a chat in five minutes.

Set it up in the dashboard, paste two lines, done. The widget is open source (MIT); cerver is the backend. You'll need a free cerver account — that's the backend.

Set up your chat → cerver-chat on GitHub