scrubjay

The app/logic for syncing Claude Code across machines. Your personal content is kept in separate repos so this one can be shared/public without leaking anything:

Repo	Role	Visibility
scrubjay (this)	scripts, hooks, docs — the logic	public-able
scrubjay-data	hosts/, settings/, claude-md/, templates/, memory/, logs/	private
scrubjay-chats	full chat transcripts (.jsonl), relayed off each machine	private

_{Diagram source: overview.dot — dot -Tsvg docs/overview.dot -o docs/overview.svg.}

Flow: scrubjay (logic) + scrubjay-data (your config) → applied into each machine's ~/.claude by claude-sync.sh. On SessionEnd a hook appends a one-line entry to scrubjay-data/logs/<host>.log and relays the session (transcript, subagents, plans) off the machine via a pluggable backend — either peer-to-peer to your own NAS (over WireGuard), or to a private scrubjay-chats repo on GitHub if you'd rather not run storage of your own. Top-level is keyed by machine so envs stay distinct and Claude can re-tailor one host's rules for another.

Where to go

• Concepts — the two-kinds-of-sync model, what scrubjay is, and what lives in the database. Start here to understand why the system is shaped this way.
• Onboarding — install on a new machine (interactive script or Claude-assisted), the repo layout, and the machine-local pointers.
• Day-to-day — the hooks that keep it hands-off, finding a past chat, and troubleshooting.
• Query the archive (MCP) — recall a past session by topic from inside a live Claude session via the read-only sjmcp server.
• Slash commands — the /sj* command reference.
• Transcripts: relay + NAS — how sessions ride peer-to-peer to your own NAS, and the two P2P paths (WireGuard / SSH).
• Reference — the by-hand command cheatsheet and environment toggles.