Harmonee's platform is built around three engines: ARIA, ALYGNMENT, and the Living Context Model. They run on the same PH22 hardware unit. They share infrastructure. But they are deliberately built as separate components, and the separation is part of why the platform works.
ARIA is the routing engine. It looks at every incoming task — classify a document, summarize a thread, draft a response, retrieve a record — and decides which model on the unit handles it. A small classification task runs on a small model. A complex multi-step analysis runs on a larger one. The routing policy is a versioned file your operations team can review and adjust. Nothing about the routing is opaque.
ALYGNMENT is the pattern-detection engine. It watches the operational signals across teams — calendar drift, document divergence, task contradiction, communication patterns — and surfaces the places where teams are stopping pointing in the same direction. It produces a weekly ranked list, not a dashboard. The output is short enough to read with your morning coffee.
The Living Context Model is organizational memory. It captures the institutional knowledge that lives in a few people's heads — who decides what, which workflow takes which path, which vendor is reliable for which job — and structures it so the rest of the team can use it. It updates itself based on observed evidence, with a human in the loop on every change.
We built them as separate components because they answer different questions and operate on different cadences. ARIA runs on every prompt. ALYGNMENT runs continuously, reporting weekly. The Living Context Model accumulates over months. A monolithic system trying to do all three would have to compromise on each.
The separation also matters for what your team can extend. The routing policy is a configuration file. The misalignment-detection signals are extensible. The organizational memory is queryable. Your team can build internal tooling on top of any of the three engines without being blocked by the others. That's the architecture we wanted, and it's the architecture the platform now is.