Research
Papers from the engine room.
Three position-and-systems pieces that describe how MingLLM works — memory, architecture, and the economics that make a personal agent viable on a laptop.
I
Memory2026
The Orb Knows
Persistent, Self-Consolidating Memory for Local Agents
Persistent memory for local agents
II
Architecture2026
One Model, Many Surfaces
A Unified Base for Voice, Web, and Code Agents
A unified base for voice, web, and code agents
III
Position2026
Against Scale
Expert Iteration Beats Frontier Compute on Bounded Agent Tasks
Expert iteration, bounded task distributions, and the end of the compute-moat thesis