* feat(cli): add --foreground flag for systemd and process manager deployments
Adds --foreground / --no-daemon to `openchamber serve` which runs the
server inline in the CLI process instead of spawning a detached daemon child.
Required for systemd Type=simple (and other process managers) that track the
direct child — the always-daemon behavior introduced in #640 broke this use case.
Also documents OPENCHAMBER_HOST (bind address) in --help, which was
implemented but never exposed to users.
* docs: add systemd service guide for VPN/LAN deployments
Documents how to run OpenCode and OpenChamber as separate systemd
user services for persistent access over Tailscale or LAN, using
the new --foreground flag and OPENCODE_HOST to wire them together.
* fix(cli): address foreground mode parity issues from PR review
- Fix Ctrl+C handling: CLI SIGINT handler now defers to server in
foreground mode; dedicated signal handlers perform graceful shutdown
and clean exit
- Restore lifecycle parity: foreground instances write PID/instance
files so status, stop, and restart can discover them
- Add deterministic --foreground --json output: emits stable startup
JSON with port, pid, url, and foreground flag before blocking
* fix(cli): tighten inline foreground behavior for restart UX and JSON-only output
* fix(cli): pass --host to foreground server, reject --json, add --quiet output
- Pass options.host through to startWebUiServer() in foreground mode so
the bind address is respected (fixes localhost-only regression from #750)
- Reject --foreground --json with a clear usage error; --json is only
supported in background (daemon) mode
- Emit resolved port on stdout in --quiet foreground mode, matching
daemon parity
- Update systemd docs to include --host 0.0.0.0 for LAN/VPN access
now that the default bind is 127.0.0.1
* fix(cli): remove duplicate OPENCHAMBER_HOST entry from help text
* fix(cli): emit restart summary before foreground serve() blocks
restart --json (and --quiet / human) with a foreground instance would
hang forever without output because serve() blocks and the post-loop
summary was unreachable. Emit the final output after stop succeeds
but before the blocking serve call — foreground is always sorted last
so all daemon results are already collected.
* fix(cli): restart stops foreground instances without re-attaching
Foreground instances are managed by a process manager (systemd, Docker,
etc.) that will restart them automatically. The restart command now
just stops the foreground instance, records the result, and exits —
no serve() call, no blocking. This makes restart --json and all
other output modes work correctly for foreground instances.
Colin Mollenhour
8dfe833faf