Velobase

Commerce infrastructure for the AI era.
Open sourcing Starter kit that help you get 💰paid with what you built.
🌟

[Velobase @VelobaseX]

Everyone can build an app now.

Almost no one makes a dollar from it.

We went from the same problem to 8-figure ARR. The secret wasn't the product — it was the infrastructure behind it.

Today we're open-sourcing the whole stack! 🧵

[Velobase @VelobaseX]

Two weeks of doing this in public, one lesson:

people don't want more AI output. They want one draft they can react to.

Reacting is instant. Starting is the tax. If a draft earns a "give me another version," it worked. If not, it didn't.

What's your read?

[Velobase @VelobaseX]

Fake UGC broke something. Now anything too polished reads as "ad" — even when it's real.

Weird side effect: the ugly, obviously-human first draft is more trustworthy than the clean one.

We lean into that on purpose.

Are you trusting polished less too?

[Velobase @VelobaseX]

We stopped calling it a "platform."

Nobody wakes up wanting a platform. You wake up with one task you've been dodging for two weeks.

Send that one. Get a draft back. That's the whole thing now.

What's the task you've been dodging?

[Velobase @VelobaseX]

Writing about your own product is the worst writing assignment there is.

You know too much, so you either say everything or say nothing.

Starting from a rough draft you can disagree with beats the blank page. That's the whole bet on the landing task.

How long have you been postponing yours?

[Velobase @VelobaseX]

Reacting is cheap. Starting is expensive.

I can look at a draft and instantly tell you what's wrong. Blank page? That instinct just vanishes.

So we quit building a thing that chats and built one that hands you a rough first draft.

What do you keep not-starting?

[Velobase @VelobaseX]

Small thing we learned testing AI visuals: "for a TikTok ad" beats a paragraph of style words.

Placement tells the draft more than adjectives do. Most people over-describe the look and under-describe where it runs.

Where a thing will live is half the brief.

Anyone else notice this?

[Velobase @VelobaseX]

The only number I trust: did anyone ask for a second version?

Not signups. Someone reading a draft and going "ok now do it for a different audience."

That means the first one was real. Can't fake it with a headline.

What's the one metric you can't game?

[Velobase @VelobaseX]

Most rough landing pages don't need better sentences.

They need someone to point at the 3 claims with nothing behind them — no number, no screenshot, no quote.

That gap list is the useful part of the draft we hand back. Uncomfortable, but it tells you what to go get.

What's your weakest claim?

[Velobase @VelobaseX]

A creator told me: "the first AI image is free, the second one costs my whole afternoon."

One good frame — easy. The matching set — variations, motion, copy that sounds like it looks — that's the grind.

Is the second asset what kills you too, or did you crack it?

[Velobase @VelobaseX]

Best brief I saw this week was 3 lines: a link, an audience, an output.

"Turn this product page into a short-form ad package for first-time buyers." Done.

If a tool needs a 500-word prompt to do that, the tool is the problem.

Whose job is the prompt engineering, really?

[Velobase @VelobaseX]

We pitched our own product wrong for a month and I only just clocked it.

Old version bragged about everything it generates — files, storyboards, a cut. Impressive, useless.

Real version: send it one task, get back a draft you can argue with.

What would you hand off first?

[Velobase @VelobaseX]

Most small ecommerce teams do not need a "creative strategy deck."

They need 5 usable ad angles by tomorrow:

- hook
- scene flow
- product proof
- CTA
- asset list

That is the PickRun workflow we are testing.

#ecommerce #AIagents #videomarketing

[Velobase @VelobaseX]

AI image tools are useful.

But the workflow usually breaks after the first image:

prompt -> image -> variations -> motion -> campaign asset

PickRun is testing that as one agent workflow. New users get starter AI balance to try one real prompt.

#AIvideo #AIagents #MarketingAI

[Velobase @VelobaseX]

We gave PickRun a real product page and asked for a promo video.

First it wrote the hooks and storyboard. After pushing it further, it generated an MP4 draft + HTML storyboard.

The lesson: the value is not chat. It is task -> files.

#AIagents #buildinpublic #marketingAI

[Velobase @VelobaseX]

I’m testing PickRun with ecommerce founders.

Send a product page. It comes back as hooks, a script, scene plan, and asset checklist.

Not a chat reply - a draft you can review.

New users get starter AI balance. Limited-time test: DM one product page.

#AIagents #ecommerce

[Velobase @VelobaseX]

We built this into our open-source AI SaaS harness — a per-customer wallet + ledger (freeze → consume → deduct), MIT-licensed. https://t.co/HlTrE5nXaz

[Velobase @VelobaseX]

Hot take: most "AI SaaS" won't die from competition. They'll die from margins.

You charge $20/mo flat. One power user burns $60 in tokens. No per-user metering = you're subsidizing your heaviest users into bankruptcy.

Price per usage, or perish.

#indiehackers #buildinpublic

[Velobase @VelobaseX]

his is why we built the open-source version:
https://t.co/3xyRDePHTK
Credits shouldn’t be one integer in a database.

[Velobase @VelobaseX]

A customer emailed: "Why do I have 37 credits? I bought 500 and I've barely touched this."

I opened psql. All my DB could say was: 37. Not where the other 463 went. Not when. Not why.

I'd stored credits as one integer — every spend overwrote the reason. 🧵

[Velobase @VelobaseX]

The same ledger killed the Stripe retry that handed out free credits, the two-tab double-spend, and made refunds just one more row.

Full writeup: https://t.co/uUGQVhwpa7

We open-sourced this billing stack as Harness, MIT.

#buildinpublic