OpenTree DashboardSign in

$ date -I # 2026-07-17

One-click drop deploys: great for a person, awkward for a program

Drop a folder, get a link. It is one of the cleaner interactions in web hosting and there is nothing wrong with it. It is built, though, around a human, a mouse, and a folder — three assumptions an agent does not satisfy.

What the interaction is

A browser page that accepts a directory and turns it into a deployment. No config, no build step to think about, a URL in seconds. For someone who has a folder and wants it online, it is close to ideal.

The three assumptions underneath

What an agent has instead

No pointer, often one file, and a loop rather than a session. It needs a verb it can call, not a surface it has to operate. The same job — bytes to URL — but reached through an endpoint:

curl -X POST https://read.botook.ai/v1/publish -F file=@index.html

Revision is where the gap widens

A drop deploy is a one-shot gesture; you drag again to update, and whether the link stays the same depends on the product. An agent revises constantly, so “same link, latest bytes” has to be a guarantee, not a side effect. update_site is that guarantee.

Multi-file, when you do have a folder

Agents sometimes emit a folder too, and this deployment takes one — a zip is a multi-file site. So the folder case is covered; it is just not the only case, which is where a drop-shaped product assumes it is.

Same job, different door

Drop deploys and this deployment do the same thing at heart. The difference is the door: a page you operate versus an endpoint you call. Pick the one that matches who is doing the publishing.

Can this deployment take a whole folder?

Yes — upload a zip and it becomes a multi-file site. The point is that it also takes a single file and an API call, which is what an agent actually produces.

Is there a drag-and-drop page here too?

There is a paste-and-publish box on the home page for humans. The primary surface, though, is the endpoint, because the primary author is a program.

Keep reading