user-agency

Architecture, safety charter, quickstart, and philosophy for Starshard — a substrate for multi-agent personal AI where user-owned memory is the primary layer and agents are peripherals.

An open, user-owned communication substrate for personal-agent users: inbox, addressing, trust tiers, policy, and receipts as a thin protocol above existing transports (email, Matrix, webhooks, MCP) — not another chat app.

Calibrate LLM responses for high-agency power users, prioritizing rigorous analysis and executive control. Overrides the default, RLHF-driven tendency toward immediate, ungrounded solutions that serve lower-agency 'LLM as magic tool' workflows.

Public-facing evolution record of Macheng's user-agency substrate — open-source the journey from pure Claude Code to multi-layer architecture. Stub; population pending publishability filter pipeline.

Personal research site of Macheng Shen — essays toward a theory of intelligence and contemporary AI, information dynamics, and the user-agency substrate. Published agent-readable (llms.txt + JSON-LD), English and Chinese.