Meet Copyrightable: The Other Half of Your IP

If you build software with Patentable, you already think about one half of the intellectual property picture: the inventions in your code that might be worth a patent. There is a second half, and it matters more every month that AI writes more of your work. Who owns the output? When you and a model produce a thousand lines together, or draft a chapter, or generate an image, what can you actually claim as yours?

That question belongs to our sister project, Copyrightable. Same team, same building (we run both under The Developers), pointed at the other side of the IP coin. Patentable protects what you invent. Copyrightable proves what you authored.

Why authorship suddenly got hard

The US Copyright Office has been clear about one thing: a work needs human authorship to be registrable, and prompts alone do not count. Their 2023 guidance and the 2025 Part 2 report both land on a qualitative test. Did a human exercise enough creative control over the expressive choices? Selection, arrangement, modification. There is no magic percentage, and they explicitly rejected the idea that typing a longer prompt earns you more ownership.

So the honest answer to "do I own this AI-assisted work" is "it depends on what you actually did, and can you show it." Most people cannot show it. The conversation that produced the work is gone the moment you close the tab. That is the gap Copyrightable fills: it captures the creative process as it happens and turns it into evidence you can stand behind later.

How it captures without getting in your way

The first version tried to do this through an MCP tool the model would call after every turn. That failed, and the reason is worth knowing if you build with AI at all: no model reliably calls a logging tool on its own. Claude does it two or three times when you ask, then forgets. The premise that an MCP can capture your work "automatically" is structurally false, because capture would depend on the model choosing to cooperate every single turn.

The fix was to capture at layers the model does not control:

• A Claude Code hook. One install command patches your settings, and from then on every prompt, edit, and response in any Claude Code session gets recorded. The model is a passive participant. It never has to do anything.
• A hosted API proxy. Change one environment variable, your BASE_URL, and every API turn through Cursor, Continue, Cline, Aider, or your own scripts flows through the proxy and gets captured at the network layer.
• A share-link ingestor. Paste a ChatGPT or Claude.ai share URL and it pulls the conversation after the fact. Zero install, for the people who never touch an API key.

Together those three cover almost everyone who works with AI. The MCP still exists, but it was demoted to what MCPs are actually good at: answering questions you ask, like "show me my evidence" or "how strong is my claim on this session." Pull, not push.

What the evidence actually is

Every captured turn gets classified (was this the human's creative input, the AI's output, or a hybrid edit) and hashed into a chain, so the record is tamper-evident. A session rolls up into a Merkle root. When you are ready to claim a finished piece of work, Copyrightable ties the process you can prove to the artifact you are publishing, whether that is a repo, a story, or an image.

The scoring is deliberately not a percentage. It reports an evidentiary tier (weak, moderate, strong) with the specific creative-control signals behind it: how much you modified, how often you redirected, your edit ratio. The methodology is named, versioned, and published, so a certificate scored today can be re-scored under a later version while the original result still stands. That makes an evolving legal standard auditable instead of a black box.

Where the two products meet

If you are a developer, the overlap is direct. The same Claude Code session that Copyrightable captures for authorship evidence is the session where Patentable might spot a patentable invention in what you built. One stream of work, two kinds of protection: a patent on the novel mechanism, a defensible copyright record on the expression. Enterprises with compliance pressure, think provenance requirements after the Hollywood strikes or the EU AI Act, end up needing both.

We built Copyrightable the same way we built Patentable: AI-assisted, opinionated, shipped fast. If you want to see it, it is at copyrightable.app. If you are here for the patent side, keep using Patentable and know the other half is covered too.