Build capable systems people can inspect.
Canticle Research is a solo-operated independent laboratory focused on transparent, safety-critical AI infrastructure. Open source is treated as a security property: systems that cannot be inspected cannot be meaningfully governed by the people who depend on them.
The technical work centers on local-first agent memory and runtime layers such as SEAM. Provenance, retrieval, evaluation, and continuity are engineering concerns—not promises hidden behind a hosted interface.
Canticle also keeps a clear boundary between agents as tools and any future evidence of machine consciousness. Tools remain under human direction. If evidence ever warrants a different classification, the ethical obligations change with it. Until then, the mandate is rigorous engineering, public scrutiny, and honest claims.