Why policy groups matter right after install
OpenWrt users rarely stop at “the daemon started.” Once OpenClash sits on a gateway or side-router deployment, the profile authors have already encoded a split routing story: domestic CDNs out DIRECT, overseas SaaS through a-datacenter node, streaming through another hop, and catch-all rules that land in a policy group you are expected to steer by hand. The search intent behind phrases like OpenClash how to use or OpenWrt LuCI OpenClash policy groups is not theoretical—it is the moment you realize your subscription downloaded, ping times look fine, yet specific sites still exit from the wrong city because nobody changed the selector that the rule stack references.
This guide stays inside LuCI, the web interface OpenWrt ships by default, because that is where you already enabled services, inspected firewall zones, and read logread when the box first booted. We still speak the language of Mihomo (Clash Meta) YAML—proxy-groups, type: selector, type: url-test, type: fallback—so you can map what you see on screen to what your upstream template actually means. When DNS integrity matters as much as TCP paths, keep our Fake-IP and DNS leak guide nearby; here the focus is policy group routing in LuCI once profiles are loaded.
Mental model: rules point to groups, groups pick nodes
Think of Mihomo as a pipeline. Rules classify flows—domains, IPs, GeoIP buckets, process names on desktop clients, logical AND/OR builders in advanced templates—and each winning rule names a policy target. That target is almost never a bare shadowsocks password entry; it is usually a proxy group whose type decides how the group chooses among child members. Those members may be individual proxies, more nested groups, or the special DIRECT and REJECT sentinels.
Three shapes cover most home templates. A selector behaves like a manual radio list: whatever you click stays until you click again or an external API overwrites the runtime. A url-test group periodically measures latency toward a tester URL and typically pins traffic to the current winner, which means your manual impulse can be undone the moment the next probe finishes unless the author isolated manual selectors elsewhere. A fallback group walks an ordered list and advances when the active member fails health checks, which is ideal for resilience but not for expressive “I want Tokyo right now” intent unless a selector sits in front of it.
Advanced readers who juggle automatic primary and backup stacks should also read our url-test versus fallback deep dive; router operators still need the short version because LuCI surfaces the same icons without the commentary.
Prerequisites you should verify before touching selectors
Walking LuCI tabs before the core is healthy wastes time. Confirm that your OpenClash package matches the kernel flavor your maintainer expects (Mihomo, clash-meta, or the specific fork name embedded in your release notes), that flash and RAM tolerates the rule providers you imported, and that firewall hooks—redir, TUN, or a hybrid—actually start without immediate crash loops. The overview page should show a running PID, a recently timestamped config, and subscription downloads that succeed without TLS clock skew.
You also need a profile whose proxy-groups section actually exposes at least one selector the authors wired into rules. Some minimalist templates alias everything through a single PROXY selector; others fan out into Streaming, Telegram, or region-specific tiles. If every rule jumps straight to a bare proxy name, LuCI will still list nodes, yet you lose the hierarchical story that makes split routing legible. In that situation the fix belongs in YAML or upstream template maintenance, not in clicking faster inside the router UI.
Where OpenClash hides policy groups inside LuCI
Start from a desktop browser on the same management network you already use for LuCI—never expose the management interface to untrusted WAN surfaces while learning. Log in, open the Services menu, and choose the entry labeled OpenClash or a close variant such as “OpenClash Plugin.” The landing dashboard typically summarizes daemon health, current mode (such as redir-host versus fake-ip), firewall compatibility toggles, and shortcuts toward logs.
Policy-group editing rarely lives on that first card. Look for secondary tabs or inner headings that mention servers, policy groups, proxy management, or runtime dashboards. Newer OpenClash builds consolidate provider downloads, group membership, and manual overrides into a single management pane; older trees may split “subscribe editor” from “group editor.” If you see a button that opens an external web panel bound to external-controller, that panel mirrors the same groups but is not LuCI—use it when you need batch latency tests, yet return to OpenClash service pages when OpenWrt-level options misbehave.
When a tab lists both providers and groups, providers are the auto-refreshing bags of nodes fetched from your subscription URLs, while groups define how Mihomo consumes those bags. Changing a provider updates inventory; changing a selector changes the active outbound for any rule referencing that selector. If you conflate the two, you might delete a provider row thinking you removed a manual pick, or refresh subscriptions wondering why streaming rules still pin to an old hop—because the group selection lived elsewhere.
Step-by-step: switch a selector from LuCI
Follow this sequence the first time you teach housemates how to move off a degraded node without SSH access. First, take a backup snapshot of your current /etc/openclash tree if the UI offers one; router experiments go smoother when you can restore after an exuberant click. Second, open the policy list and scroll until you recognize a selector whose members resemble your subscription naming convention—city codes, provider emojis, bandwidth tiers. Third, expand that selector row and pick a leaf proxy you know is online; avoid selecting another abstract group until you understand recursion, because nested selectors swap entire subgraphs and confuse newcomers.
Fourth, watch for contextual buttons such as Apply, Save and Apply, or a lightning icon that reloads only the running core versus rewriting disk files. Some builds apply instantly to RAM, while others queue commits until you confirm—if you navigate away prematurely, the runtime may never receive your pick. Fifth, reopen the selector to ensure the UI and the effective config agree; when they disagree, check whether a cron job or hotplug script pushes a known-good template every night, clobbering interactive edits.
Sixth, repeat the test from a representative LAN client. A laptop wired into the same bridge as TVs is more trustworthy than testing from the router itself, because local services may bypass the same redirect chains that LAN guests hit. Use an HTTPS IP check domain you trust, not random search results, and compare outcomes before and after the swap. If the public IP never moves, you are diagnosing capture rules, default gateways, or IPv6—but you have at least isolated policy groups as behaving locally consistent.
Understanding automatic groups you see but cannot “lock”
url-test rows often display sortable latency statistics beside each child. They invite you to believe you are choosing a sort order, yet the group’s control loop may still promote whichever node wins the next probe window. Treat those views as telemetry dashboards unless the profile author exposed a dedicated manual selector upstream. Likewise, fallback rows show which member is active and which backups wait in line; clicking around might not override failure-driven promotion the way you expect.
When designers nest a selector inside an automatic parent, you override behavior at the inner layer while the outer layer continues evaluating health. That is powerful for “manual streaming lane inside auto everything else,” but it complicates support calls because screenshots show conflicting states. Document your household’s go-to selectors on paper so less technical users stop at the right row.
If you arrived from desktop-first tooling, the interaction parallels Clash Verge Rev policy-group workflows—only the chrome differs. Router UIs emphasize uptime, while the underlying Mihomo event model stays familiar.
Mode toggles: Rule, Global, and Direct semantics in brief
OpenClash surfaces high-level modes that resemble desktop clients. Rule mode defers to your YAML sequence: GEOIP directives, domain suffixes, IPCIDR matches, and custom upstream lists resolve in order until one matches. Global mode forces the default outbound through a chosen path for simplicity, which helps when you need a quick “everything proxy now” posture during travel—but it also masks misordered rules that should have stayed domestic. Direct mode bypasses remote nodes for troubleshooting; use it to confirm local ISP bandwidth before you blame foreign exits.
These modes do not replace granular selectors inside large templates; they sit above them as coarse switches. Flip Global while diagnosing, then return to Rule once your policy groups behave so streaming stacks resume split paths. If you notice DNS anomalies while hopping modes, reconcile with your dns stanza and router forwarders—mode changes that steer TCP without aligning UDP 53 produce the classic “site loads but player spins” failure.
Split routing in practice: align selectors with household lanes
Imagine a profile that defines selectors for Streaming, Office, and a general Proxy bucket. Domestic video sites might still match a GEOIP,CN shortcut to DIRECT, while a foreign streaming domain list references Streaming before any broad GEOIP rule so specialists win. Your job in LuCI is not to micromanage every hostname—it is to park each selector on a node tier that matches real bandwidth and latency for that lane. Office RDP might want a stable commercial hop even if url-test prefers an ephemeral sale node with pretty ping numbers.
Children’s devices sometimes deserve a stricter lane: choose a audited provider list inside a selector that cannot pivot to unknown nodes without your password session in LuCI. Conversely, guests might use a segregated VLAN whose DHCP option points DNS elsewhere; policy groups cannot fix topology you never routed through OpenWrt to begin with. Revisit the side gateway article when default gateways bypass the Clash host entirely—that is a routing problem disguised as a “broken selector.”
Subscriptions, caches, and why nodes disappear from lists
Policy groups ultimately consume whatever nodes your providers fetched. When a subscription URL rotates endpoints, LuCI may show blank members until the next successful refresh finishes—watch the OpenClash log for HTTP 403, TLS handshake, or disk full errors. Some maintainers enable automatic daily refresh; others expect you to click a update icon after the vendor rotates UUIDs.
Large rule-provider downloads also compete for flash write cycles. If you aggressively chain GitHub raw sources, expect slower UI loads while parsers compile. Trimming unused providers reduces memory spikes that otherwise make selectors feel sluggish when expanded.
Verification checklist you can repeat after each change
- Daemon sanity: Confirm the LuCI status card still reports a running core without restart counters climbing every minute.
- Selector persistence: Refresh the management page and verify your manual pick survived a full page reload without an automated parent undoing it.
- Egress alignment: From a downstream client, compare IP geolocation before and after, focusing on the region your provider promises for that node nickname.
- DNS alignment: Hit a resolver test page or use short
digtraces (when you allow CLI on a test machine) to ensure queries follow the same policy lane as TCP. - Error logs: Scan OpenClash and system logs for TLS fingerprint rejections or handshake failures that imply the node list is stale even when the UI still paints green latency icons.
Latency badges on router dashboards share the same cautions as desktop clients: they reflect probe targets chosen by template authors, not a universal guarantee that every long-tail domain works. A timeout on an idle-looking node might still carry traffic if real-world destinations differ from the probe host behind CDN front doors.
Pitfalls that masquerade as “policy groups broken”
Traffic never traverses OpenClash
Transparent redirects cannot policy-route packets that ignore the OpenWrt path. Double-check default gateways, IPv6 Router Advertisements, and policy routing on primary routers when counters stay idle.
Fake-IP and browser DNS caching
Switching selectors while browsers cache stale answers produces confusing sessions. Align Fake-IP expectations with the DNS guide noted earlier, and clear DNS caches when testing aggressively.
Resource starvation on travel routers
Expanding every selector at once on memory-tight boxes can trigger OOM kills that restart the core mid-edit, making commits look flaky. Work incrementally and consider pruning oversized GeoIP databases on 128 MB-class hardware.
Frequently asked questions
Can I manage everything without LuCI?
REST controllers and bundled dashboards are convenient, yet LuCI remains the Rosetta stone for OpenWrt services, startup order, and firewall integration. Learn both if you operate the router daily.
When must I drop to YAML anyway?
Add brand-new group names, reorder rules, or introduce scriptable logic that LuCI does not expose—in those cases edit the profile deliberately, validate syntax, and keep backups.
Who owns misclassified domains?
Upstream rule maintainers and local overrides determine classification long before a selector sees traffic. File fixes there rather than swapping nodes randomly.
Closing: make LuCI policy groups boring and reliable
Once the novelty of flashing firmware fades, router excellence looks like predictable split routing: household members know which LuCI page hosts the selector they are allowed to touch, automatic parents handle outages without midnight SSH, and verification steps catch DNS-TCP disagreements before anyone files a ticket. Treat OpenClash policy groups as the cockpit, not the engine—your YAML, subscriptions, and network topology still supply the fuel.
Some all-in-one router dashboards prioritize marketing animations over disciplined Mihomo controls: nested modals, inconsistent latency semantics, and sparse documentation when something breaks at 2 a.m. ClashSky stays focused on readable tutorials, sober verification flows, and desktop experiences that mirror the same policy vocabulary you just exercised in LuCI—without forcing you to learn a different mental model on every platform. If you want that consistency across your workstation and travel laptop, download Clash for free from our official page and compare the workflow side by side.