Perplexity can appear to be unavailable even when other websites work normally through Clash. The failure may come from an unreachable proxy node, a rule that sends Perplexity traffic DIRECT, a DNS resolver that cannot resolve the destination correctly, or a mismatch between the proxy mode used by Clash and the application being tested. The fastest way to fix the problem is to isolate each layer instead of changing several settings at once. This checklist covers the browser path, Clash logs, node health, routing rules, DNS behavior, and TUN or system-proxy differences.
Fix Perplexity Not Working With Clash: Troubleshoot Access
Confirm that Perplexity traffic actually enters Clash
Start by identifying the network path used by the failing application. A browser may use Clash’s system proxy, a browser extension, a manually configured proxy, a secure DNS feature, or a separate VPN. If several of these are active simultaneously, a successful request does not prove that Clash handled it. Similarly, if Perplexity fails in one browser but works in another, the problem may be local browser configuration rather than the proxy subscription.
Open the Clash client and locate its Connections, Logs, or Traffic page. Keep the page visible, then open a new private browser window and visit https://www.perplexity.ai/. Search for a new connection involving the Perplexity domain. Depending on the browser and current application behavior, you may also see connections to authentication, static-content, analytics, or API-related domains. The important details are the destination, the matched rule, the selected proxy group, and the final connection status.
If no new Perplexity-related connection appears, the request did not reach the Clash instance you are watching. Check whether a browser extension is overriding the system proxy. Temporarily disable proxy extensions, standalone VPN extensions, and browser-specific proxy settings. On Windows, confirm that the operating system proxy points to the same local address and port displayed by the Clash client. On macOS, inspect the active network service under the system proxy settings. Do not assume that enabling a Clash system-proxy switch changes every application; only programs that respect the operating system proxy normally use that path.
If a connection appears in Clash but the browser still reports a timeout, the request has entered the core and the investigation can move to routing, DNS, or the upstream node. If the connection is marked DIRECT, the proxy rule is not being applied. If it is assigned to a proxy group but remains pending, the selected node or its remote transport may be the problem.
Connection log
Shows whether the request entered the active Clash core, which rule matched, and which policy handled the connection.
System proxy
Routes applications that respect the operating system’s HTTP or SOCKS proxy settings; it does not automatically capture every process.
TUN mode
Uses a virtual network interface to capture more traffic, but requires compatible kernel support, permissions, DNS handling, and routing.
Test the proxy node and policy group separately
A rule can be correct while the selected proxy node is unusable. Proxy providers may contain expired servers, overloaded endpoints, unsupported transport settings, or nodes that are reachable from one network but not another. Before editing the entire configuration, select a known-working policy group in Clash and test several nodes individually. Use the client’s delay test only as an initial signal: a low latency result proves that a check endpoint answered, but it does not guarantee that the node can maintain an HTTPS session to every destination.
First test the group’s current selection with a normal HTTPS website. Then open Perplexity in a fresh browser tab and watch the connection log. If the page fails only with one node, switch back to a different node without changing the rules. If several nodes from the same provider fail but nodes from another provider work, inspect the provider status, subscription update time, and remote server region. If every node fails, the issue is more likely to involve the local listener, routing policy, DNS, or the upstream service rather than one server.
Do not use a group name as proof that a proxy is selected. The group may be configured with select, url-test, fallback, or load-balance behavior. Automatic groups can change the active node after a health check, while a manually selected group remains on a node that has stopped responding. Check the actual active proxy shown in the client and, if necessary, choose a node manually for a controlled test.
proxy-groups:
- name: Perplexity
type: select
proxies:
- Fast-US
- Fast-SG
- DIRECT
rules:
- DOMAIN-SUFFIX,perplexity.ai,Perplexity
- MATCH,PROXY
This example illustrates the relationship between a domain rule and a policy group. The names are local labels and must exactly match the entries declared elsewhere in the configuration. A rule pointing to Perplexity does not work if the group is actually named Perplexity Group. Likewise, a group entry such as Fast-US must refer to a defined proxy or another supported group. Use the client’s configuration validator before applying structural changes.
When a node connects but the page remains incomplete, inspect whether the browser is reusing stale connections. Close all Perplexity tabs, clear only the site data for the affected domain if necessary, and reopen the browser. This is not a replacement for fixing routing, but it removes cached redirects, failed service-worker requests, and old connection state from the test.
Inspect routing rules for Perplexity domains
Clash evaluates rules from top to bottom. The first matching rule decides the policy, so a broad rule placed before a Perplexity-specific rule can send traffic to DIRECT, REJECT, or an unintended group. A rule for a parent domain may also affect related hostnames. Review the matched rule in the connection log instead of guessing from the order shown in a subscription editor.
For web access, begin with the primary domain and then observe additional requests made while loading the page. Perplexity may use multiple hostnames for the web application, authentication, static resources, telemetry, or API requests. You should not blindly add every hostname you see to a permanent rule. First determine which request failed and whether the hostname belongs to the service or to an unrelated browser resource.
A minimal diagnostic rule can be useful when the active configuration supports domain-based matching:
rules:
- DOMAIN-SUFFIX,perplexity.ai,Perplexity
- DOMAIN-SUFFIX,api.perplexity.ai,Perplexity
- MATCH,PROXY
The exact rule set should follow the syntax supported by the active kernel and the policy names present in the configuration. If the configuration is subscription-generated, direct edits may be overwritten at the next update. Prefer the client’s rule override, merge, or script feature when available. Save a backup before editing, and validate YAML indentation after every change.
Pay particular attention to rules such as GEOSITE,CN,DIRECT, large provider lists, GEOIP,CN,DIRECT, and custom ad-blocking providers. These rules are not automatically wrong, but their position can produce unexpected results. A domain may resolve to an address classified differently from its hostname, and a reject provider may match a resource required for page loading. The log should tell you whether the connection was rejected, sent directly, or assigned to the intended proxy policy.
| Log result | Likely meaning | Next check |
|---|---|---|
| DIRECT | A direct rule matched before the Perplexity rule. | Review rule order and broad domain providers. |
| REJECT | A block rule or provider stopped the request. | Identify the matching rule and test with a temporary override. |
| Proxy group, timeout | The request reached Clash but the selected upstream did not complete. | Test another node and verify the node transport. |
| No log entry | The application used another network path or cached content. | Check browser proxy settings, extensions, and active Clash instance. |
Check DNS resolution, leaks, and HTTPS negotiation
DNS problems can look like a dead proxy even when the selected node is healthy. The browser must resolve the hostname before it can establish an HTTPS connection. If the local resolver returns an unusable address, the request may fail before the proxy policy has a chance to work as expected. Conversely, if the browser uses its own secure DNS provider while Clash applies a different DNS policy, the observed route may not match the route shown in the Clash logs.
Review the active DNS settings in the client and configuration. Common fields include dns.enable, nameserver, fallback, fake-ip-range, and enhanced-mode. The available behavior depends on the kernel. Do not copy a DNS block designed for Mihomo into an older Clash core and expect every field to work. A configuration may load partially while unsupported features are ignored or reported in the runtime log.
dns:
enable: true
enhanced-mode: fake-ip
fake-ip-range: 198.18.0.1/16
nameserver:
- https://1.1.1.1/dns-query
- https://dns.google/dns-query
fallback:
- https://1.0.0.1/dns-query
The snippet is a diagnostic structure, not a universal recommendation. Use resolvers that are reachable on your network and supported by the installed core. If fake-IP mode is enabled, check the fake-IP filter and the client’s compatibility notes. Some applications expect real addresses, use certificate pinning, or behave incorrectly when DNS answers are synthesized. Testing with the client’s documented redirection mode or a temporary compatible DNS mode can help determine whether fake-IP handling is involved.
Browser secure DNS can bypass the system resolver. In Chromium-based browsers, review the privacy and security settings for “Use secure DNS”; in Firefox, review its DNS-over-HTTPS setting. Temporarily use the browser’s default operating-system resolver for diagnosis, or ensure that the browser’s secure DNS choice is compatible with the intended Clash DNS path. The goal is not to disable privacy features permanently, but to avoid testing two independent DNS systems without knowing which one is active.
Also check for IPv6 differences. A node or local network may provide working IPv4 connectivity while an IPv6 route fails. If the browser prefers an unusable IPv6 address, Perplexity may hang even though other sites eventually fall back to IPv4. Compare the connection log, DNS answers, and client IPv6 setting. Change IPv6 behavior only as a controlled test, then restore the setting if it is not the cause. Avoid changing operating-system-wide network parameters merely to fix one application.
Compare system proxy mode with TUN mode
System proxy and TUN mode solve different problems. System proxy changes the HTTP and SOCKS proxy settings used by compatible applications. TUN creates a virtual interface and captures traffic at a lower networking layer, allowing more programs to be routed even when they ignore system proxy variables. TUN is not automatically better: it may require administrator permission, a compatible kernel, a virtual network driver, correct DNS hijacking, and suitable route exclusions.
If Perplexity works in the browser with system proxy enabled, keep TUN disabled during the initial diagnosis. This gives you a simple path with fewer variables. If the browser does not generate Clash connections under system proxy, test TUN only after confirming that the client has permission to create the interface and that no other VPN or network filter is competing with it.
When TUN is enabled, inspect the client’s runtime log for interface creation, route installation, DNS hijack, and permission errors. A message indicating that the TUN device was created does not prove that all traffic is being captured. The browser may still use a separate VPN, a proxy extension, or a cached connection. Close the browser completely, restart the Clash client, and perform a fresh test after the TUN state is stable.
- Use system proxy mode for a clear browser-based test when the browser respects operating-system proxy settings.
- Use TUN mode when an application ignores system proxy settings or when broader device traffic must be routed.
- Do not run multiple VPN clients, transparent proxies, or TUN implementations during the same diagnosis.
- Check whether the active mode uses the same rule set and DNS policy as the system-proxy path.
- Restore the previous mode after testing if TUN causes unrelated applications or local services to fail.
Finally, restart only the components relevant to the test. Reload the configuration after changing rules, restart the core after changing kernel-level settings, and restart the browser after changing its proxy or secure DNS options. If Perplexity remains unavailable, collect the exact timestamp, selected node, matched rule, DNS mode, proxy mode, and error message. This information is far more useful than repeatedly changing subscriptions or deleting the entire configuration.