Entry 011 · 2026 · Architecture and implementation
Hermes Brain / Cortex
An auditable operating system for learning, retrieval, and agent work.
Source-linked learning runs · Deterministic publish gates · Human approval required
Hermes Brain is my private, auditable learning and operating system. Markdown, JSON/JSONL, schemas, and Git are canonical; models and agent harnesses are replaceable executors. Cortex is the retrieval and orchestration layer that makes the knowledge usable during real work.
Each deep-learning run moves through research, synthesis, challenge, evaluation, recall, human approval, and publication. Sources and claims stay linked, GBrain remains a rebuildable projection, and deterministic gates prevent an agent from publishing because a draft merely sounds convincing.
The system currently supports evidence-backed learning runs, scheduled recall, validated staging into this site's learning section, agent skills, evaluation cases, and retrieval over the approved knowledge base. The private repository holds the personal corpus; the linked public orchestration work shows the surrounding engineering direction.