We Built a Multiverse Version Control System in Rust for Parallel AI Swarms

Описание и применение

Draft — это многомерная система контроля версий, созданная на Rust, которая позволяет одновременно работать десяткам автономных AI‑агентов, устраняя проблемы дискового пространства, конфликтов и блокировок, обеспечивая мгновенное создание рабочих пространств и предсказание конфликтов в памяти.

Задача

Traditional Git’s single‑tree model breaks when 5–10 autonomous AI coding agents work concurrently on the same repository, causing disk bloat, blind collisions, and serial lock contention.

Результат

Draft achieves 21–23× faster workspace creation, 96–97% disk savings, and 10–11× faster concurrent commits compared to Git and Jujutsu, while eliminating lock contention and enabling real‑time conflict awareness for AI agents.

Рабочий процесс

• Register agent with dft agent register
• Create a new dimension dft dimension create <id>/<task_name>
• Use dft radar to view active hot zones
• Claim a file dft claim <file_path>
• Make changes, add and commit dft add . && dft commit -m "feat: ..."
• Run dft foresee <dimension> mainline to predict merge conflicts
• Converge back to mainline dft converge <dimension> mainline
• Yield the file dft yield <file_path>

Доказательства

Spawning an isolated 10,000‑file workspace completes in single‑digit milliseconds and consumes 0 KB of additional physical disk until a byte is actually modified.

Источник

10 Parallel Commits (Wall Clock): Git 1,392.36 ms, Jujutsu 537.41 ms, Draft 129.42 ms (10.8x faster).

Источник