1. Scope: reliable work calls without proxying everything
This guide is for people who already use a Mihomo-compatible client such as Clash Verge Rev, Clash Verge, or another Clash client and want to make work calls more predictable. The goal is not to force every connection through one remote node. Instead, you will identify the traffic that needs a particular route, keep ordinary local browsing and company resources on the paths they are meant to use, and test whether the change actually improves call quality.
“Split tunneling” can mean different things in different clients. Here it means using Clash rules to send selected destinations through a proxy group while other destinations use DIRECT. A rule can match a precise hostname, a domain suffix, an IP range, or—in clients and operating modes that support it—a process name. Which matching options are available depends on the Mihomo version, the profile, the operating system, and whether traffic is captured through TUN. Check the rule syntax supported by your core before relying on an example.
Zoom, Slack, and Google Meet are not single connections to one hostname. A meeting can involve sign-in, application updates, chat or signaling, media relays, content delivery networks, and integrations. Those services may use different hostnames and protocols, and providers can change their infrastructure. A rule that routes a login page correctly does not prove that the audio and video path follows the same route. Treat any hostname examples as starting points for investigation, not a complete or permanent provider list.
There is also an important boundary: using a proxy does not override company policy, a service’s terms, or regional requirements. Work accounts may be governed by an employer’s security controls, data residency commitments, or approved-network rules. Confirm that your proposed route is permitted before sending business traffic through a third-party node. Do not use a public relay for confidential calls unless your organization has approved it.
2. Why video calls are more sensitive than ordinary web browsing
A web page can often tolerate a brief pause while assets load. A live call is less forgiving. Voice and video are time-sensitive streams, and a route with high latency, jitter, packet loss, or congestion can produce robotic speech, frozen frames, or repeated reconnects even when a speed test reports a high download rate. A distant node may successfully open the meeting page yet perform poorly once it carries sustained, two-way media.
The application may also use UDP for real-time media while using HTTPS and WebSockets for account, chat, or signaling traffic. Some proxy nodes support UDP well; others do not, or perform differently under load. If media cannot use its preferred path, an application may attempt a fallback, but that fallback can be less efficient or may not work in every network. Therefore, “the app opens” and “the call is stable” are separate tests.
Desktop applications add another variable. A browser may follow the system proxy setting while a native Zoom or Slack process uses its own networking stack. An application launched before a proxy or TUN change may retain existing connections or cached DNS results. On Windows, macOS, and Linux, TUN capture can help include applications that do not honor ordinary proxy settings, but it also makes DNS handling, route exclusions, and local-network access more important.
Avoid assuming that routing every provider-owned domain through one exit is automatically better. A broad suffix rule can catch sign-in pages, static assets, or integrations that work best on a different route. Conversely, a narrow rule can miss a new media hostname and leave part of the session on the default route. The reliable approach is to observe a real session, make the smallest useful change, and retest under the same conditions.
- Latency: prefer a geographically sensible node with a consistently short round-trip time, rather than choosing only by advertised bandwidth.
- Jitter and loss: compare several minutes of call behavior, not just a single connection test.
- UDP support: check that the selected proxy path and the current Mihomo configuration support the traffic the call needs.
- Route consistency: avoid sending authentication through one path and related service traffic through an unintended, incompatible path.
3. Inventory your apps and observe actual traffic
Before editing a profile, write down what you use and what fails. Note whether Zoom is the desktop client or a browser session, whether Slack is a native app or web app, and whether Meet is opened in a browser managed by your organization. Record the operating system, the Clash client and core version, the active mode, and whether TUN is enabled. These details matter: the same domain rule can behave differently when traffic is not captured by the core.
Reproduce one issue at a time. For example, test a Zoom meeting without changing other network settings, then repeat with a chosen Clash policy. Note the start time, selected node, client mode, and symptoms: joining delay, audio drops, video freezes, or chat messages that fail to send. If the problem occurs only in a particular office, hotel, or home network, record that too. This makes it easier to separate a bad proxy route from Wi-Fi interference, a corporate firewall, or an application outage.
Use the client’s Connections or equivalent session view while signing in, joining a call, sending a Slack message, and opening a Meet session. Filter by the app process when the client exposes process information, and inspect destination domains, protocol, and the rule or policy that matched. Names such as zoom.us or slack.com may appear in some flows, but media and delivery traffic can use other hostnames. Google Meet may likewise involve Google-owned services beyond the visible meeting URL. Do not infer a complete list from one successful page load.
If your client does not show enough detail, use the operating system’s network monitor or the browser’s developer tools for browser-based calls. Keep the capture focused on the failing action and avoid publishing logs that contain account identifiers, internal hostnames, IP addresses, or tokens. For work devices, follow your organization’s logging and privacy rules. A short, redacted list of observed destinations and matched policies is usually more useful than a full configuration dump.
Make an inventory with three groups: meeting and messaging services, company resources, and ordinary personal browsing. Include your company identity provider, internal portals, repositories, file shares, and remote-access tools in the company group. Those destinations may need to remain DIRECT or use an employer-managed VPN, depending on policy. Do not assume that every domain associated with a cloud provider belongs to Zoom, Slack, or Meet; shared hosting and content delivery networks make broad guesses risky.
4. Build a small rule set and test it safely
First make a backup of the active profile or create a separate test configuration. Identify the proxy group you want to use for calls, and confirm that it is available and healthy. A group name such as Work-Call is only an example: use a name that exists in your configuration. If your profile is subscription-managed, check whether local edits are preserved on update; some clients overwrite customized rules when refreshing a remote profile.
In Mihomo-style rule lists, order matters: the first matching rule usually decides the route. Put specific, approved exceptions before broader matches, and keep your final catch-all rule at the bottom. A limited starting pattern could look like this:
rules:
- DOMAIN,meet.example.com,Work-Call
- DOMAIN-SUFFIX,approved-work.example,Work-Call
- DOMAIN-SUFFIX,internal.example,DIRECT
- MATCH,Default
The sample hostnames are placeholders, not official Zoom, Slack, or Google endpoint lists. Replace them only with destinations you have verified and are permitted to route. A real profile may use different rule types and group names. If your client supports provider-maintained rule sets, review their source and scope before adding them; a broad set can capture unrelated services. Never paste the example unchanged and expect it to route all meeting media.
For a careful first test, start with one application and one observed hostname. Add the narrowest appropriate rule, apply the configuration, and confirm in the connection log that the intended policy matched. Then test a complete call, including joining, speaking, turning video on, and sharing a screen if that is part of your normal work. Recheck chat and sign-in too, because a successful media test does not establish that every supporting service has the right route.
If an application uses many rotating hostnames, do not immediately add a broad vendor-wide suffix. Observe additional sessions and decide whether the missing destinations belong to the same service. Some clients and Mihomo configurations can match process names, but process-based routing is not universally available and may depend on TUN or platform-specific behavior. If you use it, verify the process name shown by your client and test after restarting the app; a browser process rule may affect every website opened in that browser.
Change one thing at a time. Compare the same node and the same meeting scenario before and after the change, then repeat with a second suitable node if necessary. If the connection log shows the expected rule but the call gets worse, the policy may be correct while the node is overloaded, distant, or poor at carrying UDP. Switch back to the known-good configuration rather than layering on more rules. Keep a dated note of what changed so you can undo it after a profile update.
- Back up the profile and confirm the client is using the intended Mihomo core.
- Choose one application, observe its connections, and record the matched rules.
- Add a narrow rule for a verified destination and select an approved, responsive proxy group.
- Check the connection view, then run a complete call and compare its behavior with the baseline.
- Keep, refine, or revert the change based on repeatable results—not on a single speed test.
5. Diagnose common failures without broadening rules blindly
The meeting page loads, but audio or video fails
This often means the visible web page and media connections take different routes, or that the selected path does not handle the media protocol well. Check the connection log while the call is active rather than only while loading the landing page. Look for UDP traffic, repeated reconnects, or media destinations that match the default rule. If UDP is unavailable on the selected node, test an approved UDP-capable route or compare a direct path where policy allows. Do not assume that forcing all traffic to TCP will improve a real-time call.
The browser works, but the desktop app ignores the rule
Confirm that traffic capture covers the desktop process. A system proxy setting alone may not affect every native application. If your client supports TUN, check that it is enabled, its service is running, and the relevant route is not excluded. Fully quit and relaunch the app after changing capture settings so old connections are not mistaken for new behavior. On managed devices, security software or a corporate VPN can also enforce routes that Clash cannot override.
Sign-in loops, or company tools stop working
Inspect the identity-provider and company-resource destinations separately. A broad proxy rule can cause a sign-in flow to leave through a different region or network than the rest of the session; a broad direct rule can bypass an employer-required path. Add a specific exception only after confirming the intended route with your administrator. Clear or restart the affected app’s session only when appropriate, since cached cookies and DNS results can make an old route appear to persist.
The log shows an unexpected IP or policy
Review DNS mode, fake-IP behavior, and the rule that actually matched. A hostname may resolve differently across DNS resolvers, and an IP-based rule can be less stable than a verified domain rule when provider addresses rotate. Check the final rules in their effective order, including rules imported from providers or subscriptions. If a profile update changed the result, compare the generated configuration with your saved copy and restore only the necessary customizations.
Keep a simple fallback: know how to disable the test policy, return to your previous mode, and contact your company’s IT team if a required work service becomes unreachable. A fallback is particularly important before a client update or a long meeting. Stable routing is a measured configuration, not a promise that every endpoint will remain fixed indefinitely.
6. Frequently asked questions
Should I route all Zoom, Slack, and Meet traffic through one node?
Not automatically. Start from observed destinations and your organization’s policy. A single group may be convenient, but one service can use separate sign-in, messaging, media, and delivery hosts. Test a complete session and confirm each relevant connection’s matched rule. If a broad rule catches unrelated work or local services, narrow it.
Can I use a fixed list of Zoom or Slack domains?
A published or saved list can be a useful starting point, but it may become incomplete as providers change infrastructure or assign traffic to regional services. Validate destinations in your own client logs, and review the list before relying on it. Avoid routing every domain belonging to a large shared cloud or CDN provider just because one meeting connection used it.
Is DIRECT always best for calls?
No. Direct routing can be the lowest-latency option on a reliable network, but local filtering, routing problems, or organization requirements may call for a different approved path. Conversely, a proxy can add distance, congestion, or protocol limitations. Compare the routes under the conditions where you actually work, and let your employer’s network policy take precedence.
Do I need TUN mode for the desktop apps?
Not in every setup. Some applications honor the operating system proxy, while others require traffic capture to follow Clash rules. TUN can cover more processes, but its behavior depends on the client, core, platform, and configuration. Check the connection view to see whether the app’s traffic reaches Mihomo before enabling additional capture or changing DNS settings.
Some one-click VPN tools make it easy to switch routes, but they may offer little visibility into which call connection used which path, and routing the whole device can disrupt local browsing or internal work tools. With Clash, carefully ordered rules, observable connections, and a tested proxy group let you target only the traffic that needs another route while leaving approved local destinations direct. If that level of control suits your setup, compare the available clients and choose one that matches your operating system and work policy.
Ready to get started?
Choose a compatible client, then test your work-call routes one service at a time.
Find a Mihomo-compatible client and build a work-call setup that fits your platform.
Free Download