The problem: several AI tools working on the same projects with no shared memory, repeating work, contradicting each other and losing what was already decided. The system: one workspace that Claude Code, Codex, ChatGPT, Google Gemini and Antigravity all read and write. Shared skill files, a project index, changelogs, handoffs, build notes, design files, archives and agent memory, with one set of routing rules, source authority, privacy boundaries and approval gates. Any agent picks up where the last one stopped, so the best tool for each job is used without losing context. The home desktop is the working environment; sessions are reviewed and controlled remotely from mobile. A companion application, B'Lystic AI OS (preview build), puts the same system behind a UI: project knowledge with one current authority per area, frozen context packs with checksums, agent runs tied to an exact repository state, and immutable handoffs from one agent to the next.
EvidenceClaude Code · Codex · ChatGPT · Gemini · AntigravityShared skills · index · changelogs · handoffs · archives · memoryMobile review and remote control of the home desktopB'Lystic AI OS companion app, previewSource authority + privacy rules
My RoleDesigned the structure: the skill files and how they load, the routing of each task to its authoritative source, the documentation discipline every agent follows, the handoff format between tools and the remote setup that lets me review and direct the home environment from a phone. Wrote it and maintain it as the projects grow.