---
title: "Nilkick | Launch engine on Claude | Croissant & Baguette"
description: "We built Nilkick in nine weeks: a launch-readiness scanner, a paid launch app on Claude, a Chrome extension and a public API on five Cloudflare Workers."
url: "https://croissantbaguette.com.au/projects/nilkick/"
---

# Nilkick

https://croissantbaguette.com.au/projects/nilkick/

We built Nilkick in nine weeks: a launch-readiness scanner, a paid launch app on Claude, a Chrome extension and a public API on five Cloudflare Workers.

Client: Croissant & Baguette, own product · Industry: Founders & SaaS · Services: Web Apps, AI & Workflows · Stack: SvelteKit, Cloudflare, Supabase, Stripe, Claude · Date: 2026-07 · Live site: https://nilkick.com

**Outcomes**
- **9 weeks first commit to a paid product** — free report, three paid products, extension and API
- **46 signals in the score** — 27 launch-readiness, 19 agent-readiness, none decided by an AI
- **5 Workers in production** — the app plus four job workers on Workflows and cron
- **≈$1 of AI per launch** — deep crawl, venue discovery and every draft, metered on each run
- **402 automated tests** — plus a 22-scenario simulation of the warm-up plan

## Built isn't launched.

Founders ship something good and it sits at zero. The advice they find is generic, and the real work (finding the venues that matter, reading the live threads where their users already talk, writing every post) is a grind nobody does well at eleven at night. We wanted a product of our own that did that grind with the founder keeping the final tap, and a test bed for the way we build AI systems for clients: metered spend, deterministic where it matters, a human in the loop by design.

## A score nobody can argue with.

Paste a product URL. A Worker reads the page (with browser rendering as a fallback for JavaScript shells), checks how the domain is indexed and where it is mentioned, and scores it out of 100 from 27 rules across five groups: findable, clear, crawlable, indexed, referenced. A separate agent-readiness score runs 19 checks, from llms.txt to MCP and OAuth discovery. The integer is a pure, versioned function of what was detected. No model touches it, so the same site gets the same score on refresh and the report stays shareable at a stable URL. The launch map streams in behind it: real venues joined back from a curated pool of 100, so the model cannot invent one.

## Three products on one judgment engine.

A Launch runs a deep crawl of the product, builds a brand and audience profile the founder approves, then produces a value-ranked queue of venues and live Reddit and Hacker News threads with every post already drafted, walked venue by venue. The Reddit warm-up is a 28-day plan of pure-value comments that builds account standing with no product mention, by construction. The Scout sweeps Reddit and Hacker News daily with the reply already written, and emails only on days with something new. The one rule: Nilkick never posts for you. Every submit is a human tap on the real button. Around it sits a marketing site with 57 guides, each with a markdown twin for AI crawlers, llms.txt and structured data from one content source.

## An extension and an API, for the people who want them.

A Chrome side panel docks the same queue beside the venue tab: copy each field, tap submit yourself, and the done state writes back. Copy-only by decision: no scripts on third-party pages, no autofill. For anyone who wants to drive it programmatically, a public REST API with scoped keys, signed webhooks, batch claiming and rate limits that spend model money only where a draft is asked for.

## The numbers.

Measured, not estimated. Every run meters its own model spend and timings.

## Ours, and the test bed for yours.

Nilkick is where we try things before a client pays for them: model tiers picked per task, AI spend metered on every run, deterministic scoring next to model judgment. The same stack runs the systems we build for clients.

- **SvelteKit** the app and the marketing site, with markdown twins for AI crawlers
- **Cloudflare** five Workers, Workflows for durable jobs, cron, AI Gateway
- **Supabase** Postgres and sign-in by email code
- **Stripe** credits, the subscription and the customer portal
- **Claude** Haiku for gates, Sonnet for drafting, Opus for deep ideation
- **Serper · Bright Data · Algolia** search, Reddit and Hacker News data
- **Resend** digests and transactional email
- **Vite + CRXJS** the Chrome extension