DocPull is easy to describe too narrowly.

It can pull documentation into Markdown, but that is only one use case. The larger idea is to take public web material and turn it into something local, usable, and traceable.

That material might be a docs site, a product page, a changelog, a filing, an API reference, a feed, a sitemap, or a list of URLs an agent already found. The output might be readable text, a searchable database, a saved archive, or a source pack an agent can use later.

The details matter if you are building with it. For this essay, the important part is simpler: the work should become something the user can inspect before it becomes a service.

That is what I mean by "as much as possible should be free."

Not everything. Not labor. Not support. Not the work of keeping a system alive. I mean the parts that can be copied, inspected, moved, forked, and run without consuming scarce infrastructure.

With DocPull, that means the local path has to be useful on its own. If public sources can be found without a provider, start there. If a source can be fetched locally, fetch it locally. If the result can be written as files, write files. If a run might spend money, make that clear before it happens.

This is not an argument against hosted software. Hosted software is often the right answer. Rendering difficult pages, running scheduled jobs, storing team state, handling alerts, supporting customers, and keeping a service reliable all cost real money.

Those are good reasons to charge.

The distinction I care about is where the paid boundary starts. If the cloud is doing real work, use it. If a person or system has to keep showing up, charge for it. But if the useful part can run locally and leave behind files, that path should stay open.

Files are ordinary, which is part of their value. You can open them, commit them, delete them, copy them into another project, or use them with tools I did not design. A hosted record can be more convenient, but it also asks for more trust. A local artifact gives the user something they can hold onto.

There is a related point in the current interest in hardware. The clean version is that software is being commoditized, so smart people are moving to hardware. I do not think that is literally true. Software still matters. It is how hardware becomes useful.

But the pressure behind the claim is real. Some software has become much easier to produce. Physical systems have not changed in the same way. Chips, batteries, factories, robots, data centers, deployment, and uptime still have stubborn costs.

That does not make software worthless. It makes the reusable parts of software more important as public infrastructure. If the interface layer is becoming easier to create, more of it should be available for people to inspect, copy, and build on.

That is the shape I want: the reusable part free, the responsible part paid.

The internet taught me because people left things reachable. Documentation, source code, forum threads, lectures, small utilities, old blog posts, unfinished tools. I did not know which of those would matter before they did. Nobody does.

If every useful path asks for a credit card at the first moment of curiosity, fewer people get far enough to learn from it. Fewer tools get forked. Fewer ideas reach someone outside the original audience.

"As much as possible" is doing the work in the sentence.

More is possible than we usually admit. More can run locally. More can be inspectable. More can be a file. More can wait to charge until there is a real cost on the other side.