Ping sweep + wait was the bottleneck: it blocked until all 254 pings
timed out before starting the port check. Now all 254 nc checks fire
simultaneously and we stop polling as soon as the first hit is written,
completing in ~1s instead of ~1 minute.
Instead of sequentially probing all 254 IPs in the subnet (worst case
~4 minutes), read the ARP table for devices already visible on the
network and check all of them for port 11434 simultaneously. Completes
in roughly one nc timeout (~1s) regardless of network size.
On a fresh Mac, git is a stub that requires Xcode CLT. The Homebrew
installer handles Xcode CLT automatically, so install Homebrew inline
first, then use git to clone the repo. This fixes the source failure
when step files are not yet available.
setup.sh, setup_onyx.sh, and setup_ollama.sh can all be run via
curl | bash without a local checkout. They now clone (or pull)
~/riotsecure first with inline minimal helpers, then source step
files from the cloned repo. The setup_repo step is removed from the
main flow since the clone already happens at the top.
- Extract shared helpers into steps/common.sh (colors, print_*, wait_for_user, ask_confirmation)
- Split each install step into its own sourced script under steps/
- setup.sh now sources step files instead of duplicating logic
- Add setup_ollama.sh: Ollama-only device setup (no Docker)
- Add setup_onyx.sh: Onyx-only device setup with auto-detection of Ollama host via network scan
- setup_ollama.sh prints ready-to-run setup_onyx.sh command with detected IP at completion
- Docker memory recommendation is MAX when machine has <20 GB RAM, 20 GB otherwise
- cleanup.sh and check-status.sh source steps/common.sh with inline curl fallback
- check-status.sh detects device mode (single/ollama/onyx) and skips irrelevant checks
- cleanup.sh reinstall message lists all three setup one-liners
- README updated with split-device one-liner commands
