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
Thanks for the report! Fixed in the next release: failed/slow upstream model fetches now enter a cooldown window instead of blocking every WebUI load (6cb379b), and the models list is served from the cached catalog with background refresh. Subagents list loading rides the same path. Please reopen if load times are still long after updating.
Activity
lidge-jun commentedon Jul 1, 2026
what do you mean web ui? 10100 port?
guigeng commentedon Jul 2, 2026
HTTP_PROXY & HTTPS_PROXY not in env
so the ocx can't connect openai api server and cant get models from google.
feat(net): outbound proxy support + models fetch failure cooldown
lidge-jun commentedon Jul 2, 2026
Thanks for the report! Fixed in the next release: failed/slow upstream model fetches now enter a cooldown window instead of blocking every WebUI load (6cb379b), and the models list is served from the cached catalog with background refresh. Subagents list loading rides the same path. Please reopen if load times are still long after updating.
feat(net): outbound proxy support + models fetch failure cooldown
fix(catalog,codex): derive the reasoning-rung type and scope discover…
fix(codex,usage): bind account-scoped evidence to credential generati…