Most people pay for four or five AI subscriptions and use one. Vex puts every major model on one balance — chat, code, images, video. No subscription, pay only for what you run. On that balance: a site builder that generates its own photos and background video, then audits and repairs its own code; a video pipeline where each shot continues from the last frame of the one before; a photo compositor; a CLI; and an OpenAI-compatible endpoint for tools you already use. A full site costs about $2.5
No reviews yetBe the first to leave a review for Vex AI
Maker
📌
Hi PH 👋
I built this because I was paying for ChatGPT, Claude, a video tool
and two image tools — and when I actually looked at my usage, most
months I properly used one of them. Paying for four subscriptions to
use one felt stupid.
So Vex is one balance. Chat, code, images and video all draw from it,
and you pay for what you actually run. No subscription, no separate
account per vendor. There's also a CLI and an OpenAI-compatible
endpoint, so tools you already use can run on the same balance.
What I didn't expect is that the interesting part turned out to be
what you can build once everything shares one balance.
The site builder runs several models in parallel: one writes the
page, one generates the background video, one shoots the photos. The
image legs finish first on purpose — the page model needs to see the
real pictures before it lays anything out, otherwise it designs blind
around placeholders and you get a layout fighting its own images.
Then it checks its own work. It walks the generated files looking for
broken JS, truncated HTML and references to files that were never
produced, and fires exactly one repair pass if it finds anything —
with guards so it can never loop or charge you for fixing a build
that was already fine.
A full site (code, copy, photos, background video) runs about $2 and
five minutes. A 15-second video built from three chained shots, where
each shot continues from the last frame of the one before, runs about
$1.50.
Rough edges, honestly:
- Video can't do real people's faces. The provider rejects
real-person references, so we describe the person and generate a
look-alike rather than just failing. Great for invented characters,
not for putting yourself in a shot.
- The full builder workspace is desktop-only for now — phones get a
simpler version.
- Refresh mid-comparison and you lose the answers you hadn't kept.
Drop a site idea in the comments and I'll build it today and post the
result — good or bad.
I'll be here all day. Ask me anything, including the awkward ones.