What is Klunq Proxy

Klunq Proxy is a managed gateway that lets you hand out disposable v-keys (sk-...) instead of your real provider key. It is built to pair with the Klunq widget — the embeddable page-agent at klunqlabs.com that turns any website into an AI website with one <script> tag.

The widget makes the page itself agentic: it reads content, clicks buttons, fills forms and answers questions in natural language. The proxy makes that agentic web safe to distribute: every key is capped, reset on a schedule you choose, and locked to your page.

Public beta — free models only, no payments. The proxy currently serves a small set of free models. There is no payment gate, no charges and no billing: budgets cap usage, not money. Paid top-ups arrive later under updated terms (see Terms of Service). Availability is best-effort — do not wire life-critical flows to the beta.

The proxy speaks the OpenAI API, but today it only works with the Klunq Widget. Every request must carry the widget’s watermark — without it the proxy rejects the call (see Watermarking). In the future we plan to support custom watermarks for other clients. Today: point data-base-url at https://proxy.klunqlabs.com/v1 and use a v-key from the proxy — the widget handles the rest.

Why a proxy at all

Shipping a provider key in data-api-key makes it visible in the browser inspector. The conventional fix is to build your own backend that forwards requests to the LLM and enforces limits. Klunq inverts this: keep the page agentic with zero backend, and solve key distribution centrally.

Klunq Proxy gives you:

Controllable spend — a global workspace budget plus per-key caps, both enforced
Disposable keys — revoke or rotate v-keys in seconds; safe to deploy because budgets and page-lock bound the blast radius
Page-lock — a leaked v-key cannot be repurposed for coding, math or generic chat
Curated cheap models — small open models that are enough for a page-agent at a fraction of the price
OpenAI-compatible, widget-locked — speaks the OpenAI API today, but only the widget’s watermarked prompts pass

Who is it for

The proxy is designed first for small and medium businesses that want AI on their site without operating infrastructure. You fund a single workspace budget; every visitor draws from that shared pool through their own micro-budget. Spend is limited and predictable, even if a v-key leaks.

Larger workspaces can use it the same way — your management key manages only your own v-keys.

How it fits together

Browser ─ v-key ──> Klunq Proxy ──> LLM (curated list)
                 page-lock check
                 workspace + per-key budget check

See Integration for the one-line widget change, Watermarking for the page-lock, and API reference for programmatic key management.

Status

The proxy is live for early teams. It enforces v-keys with per-key max_budget/budget_duration, a global workspace pool, page-locked watermarking (widget-only today, custom watermarks planned), and a small curated model list (gemma4 today). Spend tracking via the proxy API is operational.

Next: Quickstart — Mint your first disposable key · Concepts — v-keys & teams