Why I built an image host for AI agents (not humans)

Every image host I tried assumed a human clicking through a dashboard. Sign up, log in, drag a file, copy a link. That's fine — until the thing generating images isn't a person.

I kept running into this with AI coding agents and scripts. They produce images constantly, but to host one they'd need an account, a session, a UI. So people work around it: base64 blobs, throwaway buckets, links that rot or get region-blocked. It felt like plumbing that should already exist.

So I built PixelVault — image hosting that's API-first and agent-first. An agent, a CLI, or a script uploads an image and gets back a permanent URL in one call. No dashboard, no human in the loop.

A few decisions I made early:

• Edge-hosted, zero egress. It runs entirely on Cloudflare's edge (Workers + R2), so bandwidth doesn't turn into a surprise bill.

• Keyless when you want it. You can upload anonymously and get a URL instantly — the lowest possible friction for an agent's first call.

• Meet developers where they are. Beyond the API there's a CLI, an agent skill, a WordPress plugin, and a clipboard app in the works — all sitting on top of the same API.

It started as infrastructure I wanted for my own projects. Turns out "the boring thing that just works" is a surprisingly good product.

Would love feedback from anyone building with AI agents — what does your image/file handling look like right now?