You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
In-app browser: URL/title works but DOM inspection times out on real pages #22797
Actual: DOM calls time out / tool execution times out.
Notes
Chrome extension backend works on the same machine for the same WeChat article, so the issue seems specific to the Codex in-app browser Browser Use DOM bridge, not webpage availability.
Additional macOS reproduction with controlled A/B results.
Environment
ChatGPT/Codex Desktop: 26.825.51511 (build 7377)
macOS: 26.6.2, Apple Silicon / arm64
Bundled Browser plugin: 26.825.51511
Browser backend: Codex in-app Browser (iab)
Reproduction and timings
The behavior closely matches this issue:
about:blank
navigation: 26 ms
DOM snapshot: 14 ms
https://example.com/
navigation did not return before the Browser Use execution exceeded 45 seconds in one run
in a controlled follow-up run, goto() returned after 30.073 s
the following domSnapshot() then took another 30.063 s
URL/title remained readable
The desktop lifecycle log showed that the Browser webview attached in about 9 ms and about:blank reached DOM-ready immediately. The real page reached DOM-ready only after the fixed delay.
A fully local page at http://127.0.0.1:8765/ behaved differently:
goto(): 5.936 s
domSnapshot(): 132 ms
A/B checks
Full CDP / Developer mode
Disabled full_cdp_access_enabled, fully quit and restarted the app.
The real-page delay remained.
The original setting was restored afterward.
Network and system proxy
Direct curl https://example.com: 1.01 s
Through the configured local Clash proxy: 0.59 s
Added only example.com to the macOS proxy bypass list, leaving ChatGPT traffic on the proxy.
IAB still took 30.073 s for goto() and another 30.063 s for the DOM snapshot.
The bypass list was restored afterward.
These results do not look like normal network latency, proxy throughput, machine performance, or full-CDP overhead.
Route/session evidence
Desktop logs repeatedly contain:
No ChatGPT browser route is available for browser session <redacted>
Counts observed across separate application processes:
current fresh process: 14
earlier cold-start process: 2
previous process: 856
The previous process also recorded repeated:
IAB_TAB_SWITCH renderer timed out
The fixed 30-second DOM/control delay appears correlated with Browser Use route registration or stale session ownership rather than webpage reachability.
Could maintainers confirm whether this is a known IAB route/session regression and whether a fixed build or supported route-reset procedure exists? A useful product fix would be to rebuild the session-owned route or fail with a bounded actionable error instead of waiting for repeated 30-second timeouts.
Activity
caaaptaintop commentedon Aug 31, 2026
Additional macOS reproduction with controlled A/B results.
Environment
26.825.51511(build7377)26.6.2, Apple Silicon / arm6426.825.51511iab)Reproduction and timings
The behavior closely matches this issue:
about:blankhttps://example.com/goto()returned after 30.073 sdomSnapshot()then took another 30.063 sThe desktop lifecycle log showed that the Browser webview attached in about 9 ms and
about:blankreached DOM-ready immediately. The real page reached DOM-ready only after the fixed delay.A fully local page at
http://127.0.0.1:8765/behaved differently:goto(): 5.936 sdomSnapshot(): 132 msA/B checks
Full CDP / Developer mode
full_cdp_access_enabled, fully quit and restarted the app.Network and system proxy
curl https://example.com: 1.01 sexample.comto the macOS proxy bypass list, leaving ChatGPT traffic on the proxy.goto()and another 30.063 s for the DOM snapshot.These results do not look like normal network latency, proxy throughput, machine performance, or full-CDP overhead.
Route/session evidence
Desktop logs repeatedly contain:
Counts observed across separate application processes:
The previous process also recorded repeated:
The fixed 30-second DOM/control delay appears correlated with Browser Use route registration or stale session ownership rather than webpage reachability.
Potentially related:
Could maintainers confirm whether this is a known IAB route/session regression and whether a fixed build or supported route-reset procedure exists? A useful product fix would be to rebuild the session-owned route or fail with a bounded actionable error instead of waiting for repeated 30-second timeouts.