# Aura

A local cognitive-architecture research project that tests whether internal state can causally affect a model's generation, memory writes, tool authorization and self-repair, with every consequential action logged as an auditable receipt.

Aura is a local research runtime for testing continuous-state agency: whether an internal state can causally affect generation, memory writes, tool authorization, initiative selection and runtime repair, through code paths that leave a receipt someone else can audit. The project is explicit that it is not proof of life, personhood or consciousness, and that mechanism names in the code are not themselves evidence of what they describe.

An ontological boundary guard, enforced in code, treats loaded terms in the codebase as functional labels unless independent evidence says otherwise, and governance is treated as a design target: a lint failure or a default-approved path that bypasses it counts as a bug. Its main research programme asks whether a frozen large model checkpoint can be made to reason longer without changing its weights, and reports that this produced no measurable improvement in one preregistered test.

It runs locally on Apple Silicon and includes a test suite and reproducible proof bundles rather than asking readers to trust prose claims.

Website: https://github.com/youngbryan97/aura