Projects
NH Applied projects are developed as working systems, measured against explicit goals, and documented with the evidence needed to separate a result from an aspiration.
Host-side model execution on constrained local hardware, including functional execution of model workloads substantially larger than physical memory residency.
A public, user-controlled MCP/Agent architecture intended to let an AI client work with a computer the user owns without giving a central relay direct filesystem authority.
Reference-first investigation, provenance, bounded tests and explicit failure logging used across local AI and engineering projects.
Method
Define the question, inspect prior work and set a success boundary before optimisation begins.
Create the smallest system that can test the question without weakening the original goal.
Record failures, timings, hashes, system state and the conditions required to reproduce the result.
Separate demonstrated results from limitations and future work, then preserve the evidence.