OpenClaw picked up a useful Gateway startup fix for operators who run local assistants behind Tailscale Serve. PR #116579, titled fix(gateway): avoid false port-busy reports behind Tailscale Serve, addresses a port-detection bug that could make a healthy setup look broken.
The failure mode showed up when OpenClaw was bound to loopback, while Tailscale Serve owned the same numeric port on a tailnet address. The old fallback probe checked loopback, wildcard IPv4, loopback IPv6, and wildcard IPv6. If any bind failed, OpenClaw could report the Gateway port as busy.
That is technically overbroad. A listener on a different interface does not necessarily mean the configured Gateway endpoint is unavailable.
The User-Facing Symptom
The PR calls out an example diagnostic:
Port 18789 is already in use.
It could also suggest that process details were unavailable and point users toward installing lsof or running with more privileges. That advice was misleading in this Tailscale Serve case, because the problem was not a missing diagnostic tool. It was that the probe treated an unrelated interface conflict as if the loopback Gateway endpoint itself were blocked.
For self-hosted OpenClaw users, that kind of startup message is expensive. It nudges people into debugging the wrong layer: permissions, process ownership, or port cleanup, when the actual desired endpoint may already be usable.
What Changed
The fix narrows the check around the configured Gateway endpoint instead of treating every same-numbered interface as equivalent. If the Gateway is intended to bind to 127.0.0.1, a tailnet listener on the same port should not automatically make the local Gateway fail startup.
The PR is marked P1 with compatibility and availability risk labels, which fits the operational impact. Tailscale Serve is a common way to expose a local service safely across devices. OpenClaw should cooperate with that model without requiring operators to reshuffle ports or second-guess their tunnel.
Why It Matters
OpenClaw's Gateway is the local control point for channels, apps, browser sessions, native clients, and automations. False startup failures can make an entire personal agent look offline even when the configured route is valid.
This change improves the boring part of self-hosting: boot the service, trust the health check, and keep moving. It also makes diagnostics more honest by avoiding a generic port-busy warning when the configured interface is not actually busy.
For users who run OpenClaw over Tailscale, this should reduce a frustrating class of "it says the port is taken, but everything looks right" failures.