Automatic switching

ClashPulse switches only managed select groups you opt in: GUI Proxies > Enable group automation, TUI Proxies a, or automated_groups. Picking a node by hand removes the opt-in.

Mihomo’s url-test and fallback groups belong to Mihomo. ClashPulse never probes or switches them; url_test_interval and url_test_tolerance only set their delay settings.

Policies

Set switch_policy in [monitor], GUI Settings > Monitor, or TUI Settings y.

Fastest (lowest_latency, default) moves to a node whose median latency is lower by at least min_improvement. A fast answer does not mean a node holds long downloads.

Failover (failover) stays put until the current node fails or exceeds threshold for consecutive_bad_samples rounds.

Steady stream (steady_stream) downloads a 256 KiB HTTPS file every stream_interval through the current node and the 3 fastest others. It leaves a node after 2 of its last 5 downloads broke (reset, stalled, or cut short), for one with fewer breaks and at least 3 completed downloads. That is about 12 MB per hour per group, at most 24 MB. A node that stops answering latency probes still leaves under the Failover rule. It needs a Mihomo with pinned listener support (verified with v1.19.31).

All policies respect cooldown and manual selection, and stop switching when every candidate fails. Probes go through Mihomo’s delay endpoint, the same path as your traffic; a timeout or error is a failed sample.

Monitor settings

GUI Monitor probe policy dialog with Steady stream selected

key default meaning TUI
enabled true measure managed groups m / n
test_url http://cp.cloudflare.com/generate_204 latency probe URL u
switch_policy "lowest_latency" or failover, steady_stream y
stream_url https://speed.cloudflare.com/__down?bytes=262144 Steady stream file O
stream_interval "5m" between download rounds I
interval "1m" between latency rounds; a switch needs three i
timeout "5s" per probe o
concurrency 3 parallel probes, at most 64 c
threshold "800ms" latency that counts as bad h
alert_threshold "250ms" notify if every candidate is slower t
window_size 5 samples kept per node (2-20) W
consecutive_bad_samples 3 bad samples before failover s
min_candidate_samples 3 successes a candidate needs C
max_failures 2 failures that mark a node unstable f
min_improvement "100ms" required median gain p
cooldown "5m" between switches of one group z
jitter "10s" spread within a round w
automated_groups [] groups switched automatically Proxies a

max_failures Mihomo dial failures in a probed group also trigger an early probe, at most every 15s.

The default test_url is plain HTTP and can be intercepted. Use HTTPS if probe integrity matters. TUI T or the Monitor section sends a test notification.