A model answers. A harness is everything around it that lets the answer become an action on your machine, under your rules, with an off switch. These pages teach the vessel piece by piece, using a real one you can read: the Ormus Agent Harness.
Ten layers, forty organs. Click any piece for what it is, the scar that made it necessary, the requirements it must satisfy, and the files that implement it. Includes a self-check mode.
What a harness engineer actually decides, the eight questions every harness must answer, the myths, and the discipline that keeps the thing simple enough to own.
Every term on these pages in plain language, with a pointer to where it lives in the code. Searchable.
Ours against hermes-agent, OpenClaw, NemoClaw, rakazo and gbrain on stack and capability. Every row explains why the dimension matters and how to read the cells.
Live gap view generated from the repo at each deploy: partial and missing capabilities, what we refuse on purpose, open issues and pull requests, stamped with the tip.
Sources: docs/architecture.md, docs/synthesis.md, docs/trajectory.md, docs/comparison-matrix.md, NETER.md in the repo. Nothing on this site tracks you.