gOS is an AI computer built on macOS, with an application boundary designed to keep files, memory, tools, preferences and user judgments local unless a user authorises or configures sharing. Within macOS, FileVault protects data at rest, the Secure Enclave protects cryptographic keys and operations, and sandboxing constrains application access. These platform mechanisms support the local-first design, but they do not decide what an authorised process transmits. The user remains the human oracle because they can see context, consequences and intent that no remote model receives in full.

One machine learning on its own is worth little. A network of them can be worth a great deal, but coordination is normally bought by handing experience to a middle. gOS is designed to take a different route. When a user authorises or configures sharing, it is designed to emit a signed learning event rather than the underlying files or private memory. A valid signature binds the event to a key, making its integrity and key-based origin verifiable; it does not establish human authorship by itself.

The person at each machine lives the experience and rules on the result, supplying the claimed human context that a signature cannot prove on its own. Provenance carries that claimed source and context alongside verifiable key origin, so recipients can decide what to trust, compose or ignore. The architecture does not require a network operator to hold a view across the whole. Open Epistemic Intelligence is the research program this layer answers to.