Skip to content
Funnels for coaches, info products and DTC brands

Your agent builds the funnel. You see which path made money.

Pages written by Claude Code or Codex, sales through your own Stripe, and a funnel that learns what wins, from ad spend to revenue.

Nothing goes live without you. Agents write drafts; you publish.

A real funnel in Analytics mode: every dot is traffic moving between steps.

How it works

Write it with your agent. Sell it through your Stripe. Keep what wins.

  1. 01

    Your agent writes the pages

    Claude Code or Codex builds each page as real code over our MCP server, checks screenshots of its own work, and leaves it as a draft. Or write it in the app's editor.

  2. 02

    You sell through your own Stripe

    Every client connects their own Stripe account and the money lands there. Prices are set on the server from IDs, never in page code.

  3. 03

    It learns what wins

    Each page, component and path is scored on revenue, not clicks. A signal shows early; it only reads 'proven' after a controlled test with enough visitors and time behind it.

Funnel canvas + analytics

See the whole path, and where people leave it.

Every step of the funnel on one canvas: visitors flow along the edges, and each step shows its conversion, time on page and scroll depth.

  • Split tests compare whole paths, opt-in to checkout to upsell, by revenue per visitor
  • Time on page and scroll depth on every step, so you know if the copy was read
  • Speed on every step, next to its conversion rate
The funnel canvas in analytics mode: steps as cards with views, sales and conversion, traffic flowing along the connections.

Speed on every step

Every 100 ms costs sales. We time every step.

Amazon famously found that every 100 ms of delay cost it about 1% of sales. So every node on the canvas carries its own speed: a lab score, real visitors' load times, and a flag the moment a publish makes it slower.

  • Lab score and real-visitor load time (p75) for each step, not just the home page
  • A step that got slower is flagged on the canvas, with the before and after
  • What to fix in plain words, linked straight to the setting or the editor
A step flagged as slower on the canvas, with its Speed tab open: lab score 62, page weight, and a list of what to fix.

Ads → page

From ad spend to revenue, path by path.

Meta spend is joined to what each funnel path actually sold, upsells included. And every ad is checked against the page it lands on.

  • True path ROAS: spend, revenue and return for each ad and each split branch
  • Congruence check: flags an ad that promises one thing while its page leads with another
  • Meta Pixel and Conversions API sent from the same write that records the sale, deduplicated
The Ads page: Meta ads grouped by campaign and ad set, with spend, visitors, leads, revenue, ROAS and a congruence column.

Widgets + component library

Reuse the pieces that sell. Retire the ones that don't.

Quantity breaks, FAQs, trust badges and countdowns are widgets you configure once and place on any step. Page sections become components your agent can reuse, each scored on the visitors who actually saw it.

  • Signal first: likely helping, unclear or likely hurting, always with the visitor count
  • Proven only after a controlled test, so a lucky week doesn't crown a winner
  • Brand kit restyles every page, widget and component at once
The Widgets library: steps, benefit icons, value stack, exit intent, before and after, progress bar and more, each with where it is used.

Built by AI, in real code

No drag-and-drop. Every page is real code.

Your agent writes each page as a React file, so it can build anything a normal website can. Connect Claude Code or Codex in one line, or edit the code in the app.

  • No blocks, no templates you can't escape: quizzes, sliders, quantity tiers, custom checkouts
  • Your agent reads your brand, offers and results, and checks screenshots of its own work
  • Edit the code yourself in the app, with a live preview beside it
Claude Code
claude mcp add --transport http cashowary http://localhost:3000/api/mcp

Agents write drafts. A person reviews and publishes in the app.

What agents can do
The page editor: the page's code on the left and a live preview on the right.

Who it's for

Built for people who buy traffic and sell online.

Coaches

Application funnels, webinars and high-ticket checkouts, written by your agent in your brand.

Info products

Courses and memberships with order bumps and one-click upsells, measured to the last pound.

DTC brands

Quantity tiers, quizzes and advertorials for paid traffic, with ROAS per path rather than per click.

Agencies: run every client in its own workspace, with its own Stripe and brand.

FAQ

Questions, answered

Do I need to write code?
No. Your AI agent writes it, or you edit text, images and offers in the app. The pages are real code underneath, which is why an agent can build anything a normal website can.
Which AI agents work with it?
Anything that speaks MCP: Claude Code, Codex, Claude Desktop and others. Add our server with one command and sign in.
Can an agent publish a page or take payments?
Agents write drafts. Publishing, refunds and switching sales back on need a person in the app. The only live change an agent can make is to pause a step or funnel, for example when an offer sells out.
Whose Stripe account takes the payment?
Yours. Each workspace connects its own Stripe account and charges go straight to it. Prices are always resolved on the server.
How is 'proven' different from a normal A/B test result?
Early on you see a signal (likely helping, unclear or likely hurting) with the number of visitors behind it. It is only called proven after a controlled test passes its checks, with enough visitors and enough days behind it.
Which ad platforms are supported?
Meta today: spend, ads and creatives are pulled in and joined to funnel revenue.

Let your agent build the next funnel.

Start free, connect your Stripe, and publish when you're happy. Dangerous to your competitors. Delightful to your customers.