Ask your AI agent to create variants in your pages. RoutingCat learns what converts and routes the best converting page for each visitor.
npx routingcat initBuilt by quants who like numbers going up
Install the skills. Your agent takes it from there. The complex stuff is our problem, not yours.
RoutingCat comes with the skills for your agents to set it up on your site. Tell it what matters to track and what is the conversion metric to optimize
Your Claude Code will make it easy.
Tell your agent what alternatives to try: different images, CTA, copy, anything. Some will be better overall, and some will find their own audience.
The winner version is one prompt away.
Our models adapt to every visit with advanced machine learning to render the best variants to every visitor continuously. Every day, we use all the data to make better routing decisions that will convert better and give the weights to your server
Don’t split traffic, route it optimally.
The dashboard and CLI give rich insights about every variant, control group and generation. You and your agent can understand your performance and know what to do next.
Nothing beats an informed mind.
Your conversion rate can go up forever.
State of the art Machine Learning running 24/7 for you at agentic scale.
1/4
RoutingCat adapts to your content and visitors in real time, optimizing itself in the background. Traffic gets progressively routed to where it earns you the most. Add or remove variants at any time: routing is always ON.
2/4
Stop A/B testing every detail separately: you can optimize multiple slots (Hero, visuals, etc) at the same time, RoutingCat handles everything together efficiently.
3/4
Winning variants are one prompt away. Your agents are capable of creating, analyzing and iterating through thousands of items in parallel while you focus on your product.
› Generate 3 new customer avatars and add them as variants. Try two hero
headlines: "Start autoresearch for your conversion rate" and "Increase
your conversion rate on autopilot". And try pricing higher up, in
section 2 or 3 instead of 6.
✻ Thinking…
◆ Skill routingcat
● Read routingcat/home.ts
● Generate image public/avatars/avatar-2.png ✓ 1/3
● Generate image public/avatars/avatar-3.png ✓ 2/3
● Generate image public/avatars/avatar-4.png ✓ 3/3
● Edit routingcat/home.ts +7
hero: current
+ autoresearch "Start autoresearch for your conversion rate"
+ autopilot "Increase your conversion rate on autopilot"
avatar: avatar1
+ avatar2 · avatar3 · avatar4
+ sections: pricing6 (default)
+ pricing2 · pricing3
● Bash npx routingcat plan
home hero +2 · avatar +3 · sections new, +2
valid push will accept this
● Bash vercel --prod
routingcat LIVE home: 7 new variants, routing within about 2 minutes
✓ Production: lumen.so
Done. 7 new variants are live across 3 slots, and traffic is already
flowing to them. Expect clear signals in a few days. — 4 days later —
› How are our variants doing?
✻ Thinking…
● Bash npx routingcat status home
home · objective signup · control: 10% of visitors keep your defaults
number +41 sign-ups vs your defaults (+2.5%) · measured
hero autopilot favored +9% served 47%
autoresearch learning +1% served 18%
avatar avatar3 favored +6% served 41%
sections pricing3 learning −1% served 38%
● Bash npx routingcat show home/sections:pricing3 --by device
mobile +4% · desktop −3%
● Bash npx routingcat show home/sections:pricing3 --by time_of_day
working hours +4% · evenings −5%
Big finding: pricing in section 3 wins with mobile visitors and during
working hours (+4%), but loses with everyone else, so on average section 6
still wins (+1%). RoutingCat already routes each group to its winner:
+2.5% more sign-ups than your control so far.
"Autopilot" leads the headlines (+9%) and avatar 3 leads the avatars.
Want me to retire the weakest variants and try new ideas?4/4
Get insights into your business and each variant with comprehensive analytics and attribution. Learn what works, for who, and which channels bring you the most qualified traffic.
$100/month + $0.001 per monthly visitor over 100,000
Free under 10,000 monthly visitors.
For large scale experimentations across hundreds of products with large teams.
| extra conversion rate lift (relative) | |||||
|---|---|---|---|---|---|
| 1% | 5% | 10% | 20% | 40% | |
| extra revenue | $16,000 | $80,000 | $160,000 | $320,000 | $640,000 |
| routingcat cost | $1,000 | $1,000 | $1,000 | $1,000 | $1,000 |
| ROI | 15× | 79× | 159× | 319× | 639× |
| A/B Testing | RoutingCat | |
|---|---|---|
| Allocation | Fixed 50/50 split | Traffic flows to what converts |
| Cost of learning | Half your visitors see the loser for weeks | Losers fade as soon as the data says so |
| What visitors see | One winner for everyone | The best variant for each type of visitor |
| Time to results | Weeks of waiting for significance | Learns from day one, improves every day |
| Ideas at once | One test per page, a few variants | Dozens of variants, every slot at once |
| Workflow | Plan, launch, wait, analyze, ship, repeat | Ship or retire variants at any time |
| Who does the work | Analysts, developers, a backlog | Your coding agent, your intuition (optional) |
| Editing | Changing a variant restarts the test | Edit anytime, keep what it learned |
| After the win | Ship it and hope it holds | Keeps measuring and adapting |
| Proof of value | A lift that often shrinks after launch | Extra conversions, measured against a control group |
| Reliability | Easy to call a false winner | No winner to call: routing follows the evidence |
| Page speed | Flicker and slower pages | Zero flicker, decided before the page loads |
| Memory | Insights lost in slide decks | Every change and its result on record |
| Classic test | The only option | One setting away when you need it |
Allocation
A/B Testing Fixed 50/50 split
RoutingCat Traffic flows to what converts
Cost of learning
A/B Testing Half your visitors see the loser for weeks
RoutingCat Losers fade as soon as the data says so
What visitors see
A/B Testing One winner for everyone
RoutingCat The best variant for each type of visitor
Time to results
A/B Testing Weeks of waiting for significance
RoutingCat Learns from day one, improves every day
Ideas at once
A/B Testing One test per page, a few variants
RoutingCat Dozens of variants, every slot at once
Workflow
A/B Testing Plan, launch, wait, analyze, ship, repeat
RoutingCat Ship or retire variants at any time
Who does the work
A/B Testing Analysts, developers, a backlog
RoutingCat Your coding agent, your intuition (optional)
Editing
A/B Testing Changing a variant restarts the test
RoutingCat Edit anytime, keep what it learned
After the win
A/B Testing Ship it and hope it holds
RoutingCat Keeps measuring and adapting
Proof of value
A/B Testing A lift that often shrinks after launch
RoutingCat Extra conversions, measured against a control group
Reliability
A/B Testing Easy to call a false winner
RoutingCat No winner to call: routing follows the evidence
Page speed
A/B Testing Flicker and slower pages
RoutingCat Zero flicker, decided before the page loads
Memory
A/B Testing Insights lost in slide decks
RoutingCat Every change and its result on record
Classic test
A/B Testing The only option
RoutingCat One setting away when you need it
10% of visitors keep seeing your current site. The difference is your measured gain.
There’s no minimum. More traffic means faster learning, and unlike A/B tests, RoutingCat doesn’t wait for statistical significance before sending visitors to what works.
New variants get traffic within minutes of your deploy. Routing models retrain every day on every conversion, and results update hourly.
No. Decisions happen in your own server before the page renders: no flicker, nothing extra to load, and cached pages stay cached.
Your site keeps working. It keeps routing with the last routing table, and falls back to your current content if needed.
Today: Next.js 16+ with the App Router, hosted on Vercel. Next: Cloudflare Pages and Shopify themes. On another stack? Talk to us, we’re adding integrations fast.
Any coding agent that can run a terminal command: Claude Code, Codex, Cursor, Gemini CLI and more. RoutingCat installs a skill that teaches your agent the whole workflow. No agent? Everything also works from the dashboard and the command line.
Your coding agent, in your code: “try three headlines on the home page.” or more directed: “Try this new product picture: new_picture.png”. You review and deploy them like any other change.
The conversion you choose: sign-ups, trials, purchases, demo bookings. Revenue is tracked too.
Yes. A fixed split is one setting away, and it’s the right choice for testing prices.
Yes. Your variants show up in PostHog automatically, and conversions can come from PostHog, Stripe or your own server.
Yes. Search engines see your pages like any visitor, following Google’s testing guidelines, and bots never skew your results.
Yes:
Start experimenting with winning variants.
npx routingcat init