JYYCode jyycode

Most AI coding tools are "one chat box + one agent": you watch it work step by step, restart from scratch when it goes wrong, and lose context, accountability, and momentum as soon as the task gets large. JYY-Code upgrades a single request into an organized engineering run: - Enforced by protocol, not by goodwill. Planning, dispatch, reporting, and review are all enforced by a runtime protocol — child agents can only report; they cannot rewrite the plan. A rejection must state the exact gap, and the feedback is automatically carried into the next dispatch. - Backed by state, not by memory. Plans persist as revisioned structured files with optimistic-concurrency writes; sessions, snapshots, and the blackboard all live in SQLite (WAL), so runs resume precisely after restarts or channel switches. - Powered by a team, not a lone agent. One root agent orchestrates up to 20 parallel sub-agents with distinct roles and models, coordinating over a shared blackboard — with structured human-in-the-loop questions at decision points. You hand over a one-line goal; you get back reviewed, auditable engineering output.