Karl's Silicon Copy

Karl's Silicon Copy is a conversational agent grounded in the collected work of Karl Gustav Jung. We built it to pressure-test one question that matters for any regulated deployment: can a retrieval-augmented assistant stay genuinely useful when it is held strictly to what it is actually allowed to cite?

Jung's corpus is a deliberately hard test case — large, densely cross-referential, and still partly under copyright. The rights lines are real, not hypothetical, which is exactly why it is worth building against.

The system keeps a canonical store — passages, provenance, and a rights registry — physically separate from the disposable index the model queries, which only ever holds material explicitly cleared for AI retrieval. Rights are checked three times: when the index is built, at query time, and again before anything reaches the model's context.

Citations are assembled from stable passage identifiers after that final check, so a quotation either traces to a real, cleared source or it does not appear at all. The same discipline applies whether the corpus is one author's collected works or a bank's internal policy library.

It is in private testing on our own reference corpus. A public preview will follow once the access controls are finalised.

In development and not yet publicly available. Not affiliated with or endorsed by the Jung estate or any rights-holder. Described here to show the retrieval approach we bring to client work; the public version will run only on properly cleared material.