NeuroArch ● wip

NeuroArch is the cognitive-architecture direction of the lab: a system that doesn't just answer prompts but keeps reasoning when nobody is watching. Persistent memory, idle cognition, and beliefs that get revised over time.

Under the hood it's a market. Dozens of local models bid on cognitive work, and a router hands tasks to whoever's best for the job. Idle cycles aren't wasted — they go into research, belief recalibration, and a dream phase that crystallizes scattered notes into durable memory.

Where it honestly stands: the underlying agent substrate is already competent today — it reasons, routes, and uses tools well. NeuroArch is the architecture on top — the idle-cognition, belief-recalibration, and dream-crystallization cycles — which it doesn't yet fully run. The research is about giving a capable agent those background cycles and then measuring them, not building competence from scratch.

It's the closest thing I have to a system that's actually alive between conversations — and the next proof needed is to make those cycles measurable: memory growth, belief-revision traces, idle-cycle throughput.