Who owns the IP: the models, the prompts, and the code?

Ownership only means something if it is mechanical. The deliverable set is the source code, the fine-tuned models and adapters produced under your SOW, the prompt sets, the datasets assembled for the work, and the documentation needed to run and change all of it. It is delivered into your environment, because that is where it was built and where it already runs.

That is the same reason the ownership claim is straightforward here. The system is deployed inside your tenancy from the start, so handoff is not a migration off a vendor platform — there is no vendor platform to migrate off.

We retain no residual rights to your data or to your domain-specific artifacts. Your book, your suitability rubric, your house view, your correspondence history, the evaluation set built from your own cases, and the models tuned against them are yours and are not reused elsewhere.

The corresponding commitment on the data side is that your data is never used to train shared or third-party foundation models. Fine-tuning happens on private models inside your environment, for your use only. Ownership of the artifacts and non-use of the data are two halves of the same promise; either one alone is not worth much.

IP ownership is the question that stalls deals, and it is usually asked by people who are reading a page rather than sitting on a call. Stating the position plainly is cheaper for everyone than negotiating it at legal review. The scope of what you own is defined by your SOW, which is written against the architecture produced in Discovery rather than against a sales deck, so the list of artifacts is concrete before anyone signs.

If your own contract template has language on ownership, work product, or residual rights, the position above is the one to check it against. Anything on the boundary is worth raising in the first conversation, not the last.