Access ChatGPT Web and API in 2026: Clash Domain Rules and DNS Setup

Why a single catch-all outbound is a weak default for OpenAI

One outbound group for “the whole internet” feels simple until generative workloads mix short HTTPS requests, long-lived TLS sessions, token streaming, and cross-domain assets that do not all benefit from the same geography or congestion profile. If your default node peers poorly with the CDN fronting static files, you may blame “model latency” when the prompt never reached OpenAI’s edge cleanly. If your operating system, browser, and Clash each resolve names through a different resolver, you can burn hours rotating nodes that were never the root cause.

Clash is built for policy routing: you define proxy-groups, then write rules that send each connection to a group, to DIRECT, or elsewhere. The goal here is not to paste an unreviewed mega-list from a forum; it is to pair a compact, verifiable rule block for OpenAI-related traffic with DNS behavior that makes DOMAIN and DOMAIN-SUFFIX matchers fire during real sessions. If you are new to that model, start with the site’s Clash tutorial, then return here for an OpenAI-shaped overlay on the same YAML ideas.

How this guide complements our Gemini and DeepSeek articles

We already published a browser-centric guide for Google Gemini that focuses on Google account choreography, wide googleapis.com families, and Chromium-specific quirks. We also covered DeepSeek with an explicit split between chat.deepseek.com and api.deepseek.com. Those articles are the right references when your hostnames live under those vendors’ infrastructure. This page targets OpenAI’s own hostname family instead: public web, ChatGPT, platform keys, documentation, telemetry and feature configuration hosts used by the product, and the API origins your SDK or HTTP client calls—without recycling Gemini’s domain list or DeepSeek’s keyword focus.

If you run multiple AI products, keep separate outbound groups or clearly named child selectors so logs stay readable. Merging unrelated vendors into one bucket named AI is convenient for a day and opaque for a month. For editor-centric workflows that are not OpenAI-specific, our Cursor and GitHub split-routing guide complements this article: it emphasizes process-aware routing and repository traffic, while here we emphasize hostname clarity for OpenAI’s web and API surfaces. For another vendor’s web-versus-API split, compare with our DeepSeek split-routing guide.

What you are really routing: web chat, platform, docs, and API origins

OpenAI sessions typically touch more than one name, and the vendor can add hosts after product updates—so treat the following as a baseline to confirm in your own logs, not as an eternal static truth.

• Public marketing and policy pages: openai.com, www.openai.com, and sibling hosts used for announcements and legal content.
• ChatGPT in the browser: the application commonly lives at chat.openai.com (paths may change; Clash matches the hostname in SNI-aware flows).
• Developer platform and API keys: platform.openai.com when you manage projects, keys, usage, or billing in a browser tab.
• HTTPS API for Chat Completions and related endpoints: api.openai.com for many REST-style calls; newer or specialized products may introduce additional API hosts—always verify the base URL in the current official documentation for the API product you use.
• Authentication and session support: flows may involve auth.openai.com or other identity-related names; reproduce failures with logging enabled rather than guessing.
• Operational and configuration fetches: some clients contact hosts such as features.cfg.openai.com or similarly named configuration endpoints; missing these can produce “stuck loading” symptoms that look like model outages.

Static assets, analytics, error reporting, and third-party embeds can add more names. Do not guess: reproduce the failure with logging enabled, read the Server Name Indication or destination names your machine actually uses, then extend the list surgically. If you maintain rules through remote providers, see our custom rules tutorial for merge order and how subscription refreshes can erase personal tweaks.

Web streaming, HTTP/2, and why “API works but chat stutters”

Browser chat often keeps a connection open while tokens arrive. Command-line API calls may complete in one request-response cycle. That difference matters when your exit shapes traffic aggressively, when UDP or QUIC paths diverge from TCP, or when a middlebox resets idle streams. When symptoms split along browser versus CLI, compare capture mode first: system proxy settings do not always intercept every helper process, while TUN mode can surface more flows at the cost of complexity. For transparent capture background, read our TUN mode deep dive.

IPv6 and dual-stack gotchas

On networks that advertise IPv6, your OS may prefer AAAA records. If your proxy path handles IPv4 and IPv6 asymmetrically, you can see intermittent failures that correlate with switching between Wi-Fi and tethering rather than with OpenAI itself. When debugging, note whether log lines show v4 or v6 destinations and whether you need explicit IP-CIDR6 DIRECT lines for local ranges, mirroring what you already do for RFC1918 IPv4 space.

Design outbound groups: one bucket or two for OpenAI?

Before editing rules, define proxy-groups entries you can aim at. A single group named OpenAI is enough when the same exit satisfies both web and API. Two groups—OpenAI-Web and OpenAI-API—help when you want different regions, different failover policies, or stricter latency targets for programmatic calls while keeping the browser on a more stable path.

Prefer select when you want manual control, url-test or fallback when you want automatic rotation. The nodes must actually complete TLS to OpenAI endpoints without broken certificate inspection or half-configured IPv6. For scheduling mechanics inside YAML, our proxy groups guide explains selectors, health checks, and nesting without tying the story to a single vendor.

Keep these groups separate from a generic Proxy catch-all so your logs answer a simple question: when ChatGPT failed, did OpenAI traffic hit the intended policy name? If the answer is no, fix capture or rule order before you chase the fifth node in a list.

Domain rules: conservative matchers and precedence

Clash evaluates rules top to bottom; first match wins. Place LAN exclusions, private ranges, and other high-confidence DIRECT lines before vendor-specific matchers. Then add OpenAI-related names with DOMAIN for exact hosts and DOMAIN-SUFFIX only when you understand the blast radius—DOMAIN-SUFFIX,openai.com,OpenAI is simple and broad; it may also route subdomains you did not intend if OpenAI later introduces edge services you wanted on DIRECT.

Developers who run scripts on servers without a browser still benefit from the same idea: your HTTP client resolves api.openai.com; if resolution is poisoned or split-horizon, the TLS handshake never reaches the policy you wrote. That is DNS-first debugging, not “try another exit in the same city.”

YAML skeleton: LAN first, then web and API hosts

Assume your profile already defines proxies and a group named OpenAI. The fragment below is illustrative: adapt names, merge with your provider template, and verify hostnames against your own capture.

# Local and loopback first (adjust to your network)
IP-CIDR,192.168.0.0/16,DIRECT
IP-CIDR,10.0.0.0/8,DIRECT
IP-CIDR,172.16.0.0/12,DIRECT
IP-CIDR,127.0.0.0/8,DIRECT

# OpenAI / ChatGPT — verify in YOUR logs after vendor updates
DOMAIN,api.openai.com,OpenAI
DOMAIN,chat.openai.com,OpenAI
DOMAIN,platform.openai.com,OpenAI
DOMAIN,auth.openai.com,OpenAI
DOMAIN,features.cfg.openai.com,OpenAI
DOMAIN-SUFFIX,openai.com,OpenAI

# Remaining traffic follows your profile (GEOIP, MATCH, etc.)
# MATCH,Auto

If you split web and API across two groups, duplicate the DOMAIN lines with different targets—OpenAI-Web for chat and platform hosts, OpenAI-API for api.openai.com—and keep the order consistent with your operational intent. The YAML is not magic; it is a deterministic decision list applied to each connection Clash sees.

RULE-SET workflows for teams

Individuals can maintain a short inline block. Teams often prefer a RULE-SET remote file or internal Git snippet so reviewers can diff changes. Meta-class cores support rule providers; the maintenance challenge is the same as inline YAML—document ownership, pin update intervals thoughtfully, and ensure your personal overrides survive subscription merges. When a provider refresh reorders rules, rerun your short log check instead of assuming “OpenAI broke.”

Keep machine-readable comments in a separate changelog if your provider strips comments. Humans forget why platform.openai.com was added; future you will not remember unless you wrote it down outside the auto-generated blob.

When you publish a shared ruleset internally, version it like any other config artifact: semantic tags, a short README that states assumptions (“desktop browsers only,” “includes API for CI runners behind corporate proxy”), and a rollback path. Code review for YAML is boring until the day it prevents an incident; treat AI vendor rules with the same seriousness as firewall ACLs that touch production egress.

Where hand-written rules beat giant community lists

Community-maintained “AI rulesets” can save time, but they also age unevenly: one contributor’s REJECT line for analytics might block a telemetry hostname your client now requires, or a stale IP-CIDR entry might send traffic to the wrong continent after the vendor renumbers. For OpenAI, logging first and adding lines beats importing ten thousand lines you cannot explain. If you do import a remote set, fork it, pin the URL, and schedule periodic reviews—automatic updates without human attention are how surprises compound.

DNS: the hidden half of correct domain rules

Misconfigured DNS makes split routing look “random.” In fake-ip modes, Clash maps domain queries to synthetic addresses internally; that is elegant until a browser uses a different encrypted resolver and caches divergent answers. Symptoms include intermittent TLS failures, endless loading spinners, and “worked until reboot” behavior.

Align deliberately. If applications use DNS over HTTPS directly, those queries may bypass assumptions your DOMAIN rules rely on, because the core observes an IP connection without the domain context you expected. Mitigations are practical, not ideological: route known DoH provider hostnames through the same policy as the app, steer DoH to a resolver you control, or accept IP-based classification and document the trade-off. The objective is consistent name-to-policy mapping across the processes you care about.

When the web UI loads but API calls from a terminal fail (or the reverse), compare which resolver each tool uses. IDEs and language runtimes frequently ignore OS proxy settings unless configured; they may still honor HTTP proxies when set, but DNS might be OS-level or library-level. Uniform debugging beats swapping exits blindly.

Poisoned answers, captive portals, and restrictive networks

Not every strange DNS response is malware. Hotels and coffee shops return synthetic answers until you authenticate. Some enterprise filters categorize AI domains inconsistently. If OpenAI fails only on one physical network, test a phone hotspot before you rewrite YAML. Correlation saves time.

Split DNS versus “one resolver to rule them all”

Power users sometimes configure different upstreams for domestic versus foreign names. That can work well when documented, but it increases cognitive load for family machines. Pick a strategy that matches who operates the computer: a disciplined single-path resolver behind Clash is often easier to support than three parallel experiments fighting each other.

Another subtle failure mode is negative caching: a transient NXDOMAIN or SERVFAIL during a flaky network moment gets cached by an intermediate layer, and every subsequent attempt “proves” the hostname does not exist until a TTL expires. When ChatGPT intermittently fails with name-resolution errors in one application but not another, flush caches methodically—browser, OS stub resolver, and any security product that implements its own mini-DNS—and retest on a clean network before editing Clash.

TLS SNI, ESNI/ECH, and “why my DOMAIN rule did not match”

Most user guides implicitly assume visible SNI hostnames. Encrypted Client Hello and related privacy features change how much a local proxy core can infer without additional configuration. If your client stack enables aggressive privacy modes, you may see more IP-only flows hitting GEOIP or final MATCH lines than you expect. When that happens, either accept broader IP-based policies with documented risk, adjust client settings for controlled debugging, or route known CDN IP ranges with explicit caution. Domain rules express intent about names; if names disappear from the wire, policy must adapt.

System proxy versus TUN for browsers, terminals, and containers

System proxy mode is usually the gentlest first step on desktops: browsers pick it up, and many GUI clients integrate cleanly. Yet terminals, language package managers, and Docker workloads may not use the same environment variables. TUN mode raises capture rates at the cost of occasional conflicts with other VPN products or corporate agents.

A practical sequence is: confirm Clash loads the profile you think it does; reproduce a minimal OpenAI action with logs open; if connections never hit the core, escalate capture rather than adding more domain lines. Disable competing full-tunnel VPNs during tests—two layers arguing over routes produces “half the internet works” reports that waste weekends.

CI runners, cloud shells, and headless API clients

Continuous integration environments often lack a Clash sidecar entirely. The split-routing lessons still apply conceptually—predictable DNS and egress—even when implementation shifts to corporate HTTP proxies or allow-listed NAT gateways. If you develop locally with Clash but deploy to a locked-down server, document the difference so HTTP 401/403 errors are not misread as routing bugs.

Local scripts that read HTTPS_PROXY may still perform DNS resolution through libc before the CONNECT tunnel forms; if resolution fails, no proxy rule ever runs. Exporting ALL_PROXY or teaching libraries to use a SOCKS5 front-end can change that story, but the fix is library-specific. When helping teammates, share a minimal reproduction—ten lines of Python or curl with verbose flags—rather than a screenshot of a GUI.

API keys, logs, and operational hygiene

Developer guides rightly stress never committing API keys. Operational reality also means not pasting keys into random “test” chat windows and not leaving debug logging enabled on shared machines where logs aggregate to a vendor SIEM. Clash logs can include destination hostnames and timing metadata; depending on verbosity, they may surface enough context to reconstruct usage patterns. Treat log retention like any other sensitive artifact: rotate, redact, and scope access.

When rate limits or quota errors appear, exponential backoff is table stakes. Split routing does not exempt you from polite client behavior—stable proxy paths make it easier to accidentally hammer endpoints from a long-running loop. Instrument your jobs with request IDs and clear error classification so you know whether a 429 is a quota story, a regional capacity story, or your own bug in retry logic.

Mobile browsers and companion experiences

Phones switch radios aggressively; DNS caches and “Wi-Fi assist” style features can route subflows in ways desktop users rarely see. If you run Clash-class clients on mobile, verify hostnames on the device itself. Do not assume a working laptop profile transfers one-to-one when the mobile client uses per-app VPN semantics or split tunnel lists managed by the OS vendor.

Background refresh may delay when a chat UI reconnects after sleep; that can look like “ChatGPT is down” when the actual issue is power management starving network tasks. Before rewriting rules on mobile, compare behavior on a stable charger-connected session with background restrictions lifted.

Verification workflow you can repeat in about a minute

First, confirm the active profile and that local overrides survived any subscription refresh. Second, open logs and run a minimal web test: load ChatGPT, send a short prompt, wait for completion or failure. Third, run a minimal API test from the same machine—curl or a tiny script—to api.openai.com using your real key in a safe environment. Fourth, note which rule matched and which outbound group handled each flow. Fifth, only then rotate nodes inside OpenAI if throughput or loss remains suspect.

When authentication misbehaves, widen the window: account-related hosts might still hit DIRECT because an earlier rule swallowed traffic. When streaming stalls, check UDP/QUIC paths and MTU before you assume model saturation.

What to record when something regresses

Capture the profile version, core flavor, capture mode, three example destination hostnames from the failure window, and the network type. Browser updates and OS “secure DNS” toggles are frequent silent variables. A short, structured note turns “it broke again” into a solvable diff.

Symptom cookbook: likely causes before you blame the model

• 401/403 on API calls while the web app works: keys, billing, or organization policy are primary suspects—verify credentials independent of Clash. If only CLI fails, check whether the terminal uses a different proxy or DNS path than the browser.
• Chat loads but responses never start: inspect whether streaming endpoints are blocked by a premature REJECT or an upstream that strips long-lived connections; compare with a short non-streaming request.
• Timeouts only on one network: correlate with captive portals, IPv6 preference, or carrier-grade NAT; compare hotspot versus office Ethernet.
• Everything fails after a subscription update: diff merge order—provider templates sometimes insert broad GEOIP or early MATCH lines that bypass your OpenAI block.
• “Works in incognito, fails in normal profile”: suspect extensions that rewrite headers, force alternate resolvers, or inject corporate inspection certificates differently per profile.
• Intermittent TLS handshake errors: check system clock skew, custom root stores on security appliances, and whether a different exit presents a captive portal HTML page instead of a certificate chain.

Use the list as orientation, not scripture. Logs remain authoritative; cookbooks reduce the search space so you do not spiral through unrelated forum threads late at night.

Making overrides survive subscription churn

Most people import remote profiles. Auto-updates can replace rules wholesale. Prefer client features that prepend or append user snippets, or maintain a local merge file you control. After every refresh, rerun the short verification sequence. Treat it like a smoke test for infrastructure you rely on daily.

Performance tuning without fooling yourself

Latency to inference endpoints is only one variable. Thermal throttling, background sync, and aggressive browser extensions can mimic network stalls. Before you add a sixth domain guess, close heavy tabs, disable a suspect extension briefly, and retest. Separate application slowness from path slowness; Clash only addresses the latter directly.

Privacy, terms, and realistic expectations

Routing changes path selection; it does not replace compliance with service terms, workplace policies, or regional regulations. Corporate devices may forbid split tunneling. This article assumes you configure a machine you own or legitimately administer.

Prefer the site’s download page for maintained clients; treat upstream GitHub repositories as transparency and issue tracking, not necessarily as the first click for every installer decision.

Putting it together

Stable ChatGPT and OpenAI API access with Clash in 2026 is less about secret host lists and more about a tight loop: observe names in logs, encode them into focused domain rules aimed at dedicated groups, align DNS with capture mode and DoH reality, and prove matches before swapping nodes. Compared with global proxy toggles, that approach keeps unrelated traffic on sensible paths, makes web and API failures easier to separate, and survives vendor infrastructure churn if you treat lists as living documents.

For a parallel example aimed at Google’s browser ecosystem, see our Google Gemini rules and DNS guide. When you are ready to install or standardize a maintained Clash Meta-class client, walk through our Clash tutorial and use our download page as the primary path—Download Clash for free and experience the difference.