1. Scope: route work traffic deliberately, not by habit
This guide is for people who already use a Clash client with a Mihomo-compatible profile and want a more dependable work-from-home setup for Zoom, Slack, and Google Meet. The goal is not to push every connection through a proxy. It is to choose an appropriate path for calls and collaboration while keeping local services, company resources, and ordinary browsing on the connections that make sense for them.
A video call is different from loading a web page. Calls are sensitive to latency, jitter, and packet loss; a relay that makes a page load reliably may still add enough delay to make conversation feel awkward. Chat and file uploads have different needs again. Slack may keep long-lived connections open for messages while fetching images or files from separate hosts. Meet and Zoom can use a mix of HTTPS signaling and real-time media transports. The exact endpoints depend on the app version, account, operating system, network, and features in use, so a short copied hostname list is not a substitute for observing your own traffic.
Before changing a profile, check your organization’s security and network policies. A proxy can change the apparent source region of a connection and move work traffic through infrastructure operated by someone else. Do not send company calls, messages, recordings, or documents through a third-party node unless your employer permits that path. Keep credentials and subscription URLs private, and treat any remote node as infrastructure you do not control.
For a baseline, decide which outcome you want for each app: direct access, a chosen proxy group, or the organization’s approved VPN. A sensible starting plan is to keep corporate intranet and local network traffic direct, route collaboration app traffic consistently, and leave unrelated web browsing on your normal policy. Measure the result during a real meeting rather than judging the configuration from a successful sign-in screen.
2. Why calls need more than a browser-level proxy switch
A browser proxy setting only affects connections made by that browser. The desktop Zoom client, Slack application, and Google Meet in a browser may each use different network stacks, DNS behavior, and transport fallbacks. A desktop app can open a media connection that does not follow a browser’s manual proxy setting; a browser may use system proxy settings for HTTPS while its real-time media path behaves differently. That is why “the website opens” does not prove a call is routed the way you intended.
Real-time audio and video commonly use UDP when the network allows it, with fallback behavior when UDP is unavailable or blocked. A proxy node or network path optimized for ordinary TCP browsing might not handle that media path well. Conversely, a route that provides a reachable signaling endpoint does not guarantee low-latency media. Watch for the actual symptom: a call that joins but has one-way audio, a frozen video feed, repeated reconnection, or a large delay may point to different parts of the path.
Clash rules select policies based on information the core can see, such as domain names, IP ranges, process names where supported, and rule order. They do not understand that a particular packet is “important because this is a meeting.” With domain rules, the application’s signaling and media traffic may not all use the same hostname. With process rules, the application may spawn helper processes or use a shared browser process. A robust setup therefore combines a small, verified set of app rules with careful testing and a clear fallback policy.
Keep rule order in mind. Mihomo evaluates rules from top to bottom and applies the first matching rule. A broad rule such as GEOIP, DOMAIN-SUFFIX, or a catch-all MATCH can take effect before a more specific collaboration rule if it appears earlier. Put exceptions and deliberate app routes above broad regional or final rules. Also remember that a rule can only match what the core observes; DNS mode, TUN settings, system proxy integration, and application behavior can all affect visibility.
3. Inventory Zoom, Slack, and Meet traffic before editing YAML
Start with a repeatable test on the network where you actually work. Record your current Clash mode, DNS mode, active profile, and the name of the selected policy group. Then test one app at a time: sign in, join a test call, send a message, and open a file or shared link. Avoid testing all three apps at once, because the resulting connection log becomes harder to interpret.
In a Clash client, open the connection or traffic view while reproducing the behavior. Look for destinations, matched rules, selected policies, connection duration, and whether the connection is TCP or UDP when that information is available. If a call fails, note the time and compare the entries immediately before and after the failure. On desktop systems, operating-system tools can help identify which process opened a connection; the process name may be more useful than guessing from a generic cloud hostname.
For browser-based Meet, inspect the browser’s network activity for page and signaling requests, but do not assume the browser developer tools will show every media flow. For Zoom and Slack desktop clients, use the Clash connection list and system process information together. A destination that appears to belong to a large content-delivery network may serve many unrelated services; do not route an entire provider domain through a special policy just because one test connection appeared during a call.
Build a small inventory in a private team note. Record the app, version, operating system, observed hostname or process, matched rule, and what happened in the test. Mark uncertain entries as provisional. Vendor endpoint lists and application behavior change, and large services may use regional or shared infrastructure. Recheck your notes after client upgrades, policy changes, or a move between home Wi-Fi, a hotspot, and a corporate network.
| Work surface | What to verify | Useful test |
|---|---|---|
| Zoom desktop | Sign-in, meeting control, media path, and whether the app uses UDP or falls back | Join a test meeting and check audio in both directions |
| Slack desktop | Message connection, workspace sign-in, file access, and app update traffic | Send a message, open a thread, and download an approved test file |
| Google Meet in a browser | Browser proxy behavior, call signaling, and separate media connectivity | Join a test call, share a screen, and confirm both participants can hear and see |
| Company services | Internal DNS, intranet, SSO, and approved VPN requirements | Open a company resource before and after enabling Clash rules |
4. Build policy groups and rules that are easy to reason about
Use policy group names that describe the decision, not a temporary server location. For example, keep a group for collaboration traffic, another for ordinary proxy use, and a direct policy for destinations that should remain on the local connection. The group can let you choose a suitable node without rewriting each rule. If you already have provider groups in your profile, reference the existing names rather than inventing names that do not exist in your configuration.
The following fragment illustrates the structure, not a complete drop-in profile. Replace the example group names with groups that are actually defined in your active configuration. Add only domains you have verified from logs or official documentation. A broad suffix match can affect unrelated services, so start narrowly and expand only when a real app feature is missing.
proxy-groups:
- name: Work-Apps
type: select
proxies:
- DIRECT
- YOUR-EXISTING-PROXY-GROUP
rules:
# Keep company and local destinations on approved paths.
- DOMAIN-SUFFIX,corp.example,DIRECT
- IP-CIDR,192.168.0.0/16,DIRECT,no-resolve
# Add verified collaboration domains above broad rules.
- DOMAIN,verified-service.example,Work-Apps
# Keep your existing regional and final rules below these entries.
- MATCH,YOUR-EXISTING-DEFAULT-POLICY
Do not paste that example over a subscription-managed profile without checking how your client handles updates. A provider may regenerate the configuration, overwrite edits, or expose a ruleset section designed for local overrides. Prefer the client’s supported merge or override mechanism when available, and keep a backup so a syntax mistake does not leave you without a working profile. Validate YAML indentation and confirm that every referenced policy exists before applying changes.
For Slack, Zoom, and Meet, use observed hostnames rather than a guessed list of every domain associated with their vendors. Authentication, static assets, analytics, file storage, and media may use different services. A single organization domain could also host both work-critical and unrelated products. If a domain rule catches too much, remove it and test a narrower match or a process-based approach supported by your core and operating system. Do not add an entire shared CDN to the work group just to make one login screen load.
Keep local and corporate exceptions above broad routing rules. Company SSO may redirect through an identity provider, and an internal portal may resolve to a private address even when the page name looks public. If your company requires its own VPN for internal systems, follow that requirement rather than using Clash to replace it. For local printers, NAS devices, and development services, confirm that the profile’s private-network behavior is intentional; a direct route should not accidentally expose a service on an untrusted network.
5. Test the result, then tune one variable at a time
After applying the profile, fully quit and reopen the affected app if it may cache proxy settings or keep old connections alive. Check that the new rules appear in the active configuration, then reproduce the same test you used for the baseline. For a video call, verify microphone and speaker audio in both directions, camera quality, screen sharing, and stability over several minutes. For Slack, check sign-in, message delivery, and a permitted file action. A successful home page load is not a complete test of any of these workflows.
Compare direct and proxy routes only when your employer’s policy allows both. Change one setting at a time: first the selected policy, then a specific rule, then DNS or TUN behavior if needed. If the call becomes worse after selecting a distant node, try a closer approved route or test direct access. Do not assume that a higher bandwidth speed-test result means a better meeting; latency variation and packet loss often matter more for interactive media than peak download speed.
When the symptom is a sign-in loop, inspect authentication and SSO destinations as well as the main app domain. When chat works but media fails, check whether the media flow uses a transport or process that your rule does not capture. When the app connects but company resources stop resolving, review DNS settings and corporate routes before adding broad exceptions. If only one network causes trouble, compare Wi-Fi, hotspot, and VPN conditions; the issue may be a router, firewall, or network policy rather than a Clash rule.
Keep a rollback path. Save the last known-good profile, note the rule you changed, and avoid editing a live work configuration immediately before an important call. If a profile update changes a group name or rule order, revalidate your overrides. A small, documented set of decisions is easier for a teammate to audit than a long list of unexplained domains. Revisit the inventory periodically, but do not refresh rules just to chase every hostname in a connection log.
Some one-click VPN tools are convenient, but their all-or-nothing routing can send local browsing and corporate services through the same distant exit as a meeting, while application-specific proxy settings may miss desktop media traffic altogether. Clash gives you explicit policy groups, ordered rules, and connection logs, so you can inspect and adjust the path for verified work traffic while preserving direct routes where appropriate. If you want to set up a compatible client and continue with a broader configuration guide, start with the download and tutorial pages below.
Ready to get started?
Choose a compatible client, then apply and test work-app routing with a profile that fits your network.