1
00:00:00,000 --> 00:00:05,440
Imagine switching from your personal
ChatGPT account over to your company enterprise

2
00:00:05,520 --> 00:00:10,320
login, thinking you are completely
isolated, but under the hood,

3
00:00:10,336 --> 00:00:15,040
your terminal daemon is still holding open
live connections and model permissions

4
00:00:15,088 --> 00:00:16,400
from your personal session.

5
00:00:17,497 --> 00:00:18,297
Wait, hold on.

6
00:00:18,757 --> 00:00:19,337
You mean like...

7
00:00:19,897 --> 00:00:23,917
background tasks running under the wrong
credentials entirely?

8
00:00:24,333 --> 00:00:25,533
Exactly that.

9
00:00:25,720 --> 00:00:31,053
In the Codex CLI app server architecture,
the background daemon was caching things

10
00:00:31,101 --> 00:00:36,893
like model catalogs, active WebSockets,
and remote control sessions globally,

11
00:00:36,933 --> 00:00:40,013
without strictly tying them to your active
auth identity.

12
00:00:40,442 --> 00:00:46,782
Wow, so if you toggled between personal
ChatGPT Pro and an enterprise

13
00:00:46,862 --> 00:00:51,282
OIDC login, those persistent remote
control handles could just...

14
00:00:51,842 --> 00:00:52,422
sit there.

15
00:00:53,422 --> 00:00:58,202
According to recent release notes,
switching accounts now invalidates remote control

16
00:00:58,322 --> 00:01:03,422
sessions, cached WebSocket state, and
model catalogs belonging to the previous

17
00:01:03,502 --> 00:01:03,982
identity.

18
00:01:04,375 --> 00:01:09,735
Yeah, that fix landed in Codex zero point
one fifty five point zero across pull

19
00:01:09,806 --> 00:01:14,975
requests forty three nine zero six and
forty three eight forty four.

20
00:01:15,043 --> 00:01:20,855
Basically, model catalog caches are now
compound keyed by provider and auth identity

21
00:01:20,926 --> 00:01:21,575
together.

22
00:01:21,735 --> 00:01:27,255
And when you trigger an account logout or
switch, it forces an immediate teardown.

23
00:01:28,512 --> 00:01:34,072
Which means developers can finally stop
running pkill codex in their terminal every

24
00:01:34,132 --> 00:01:35,632
time they switch client projects.

25
00:01:36,052 --> 00:01:38,012
That was such a messy workaround.

26
00:01:38,375 --> 00:01:43,895
Right, um, the important operational
caveat now is that if you do switch identities

27
00:01:43,935 --> 00:01:47,815
while a background turn is actively
running under the old profile,

28
00:01:47,875 --> 00:01:52,375
Codex will abort that turn cleanly instead
of letting it execute with mismatched

29
00:01:52,423 --> 00:01:52,935
authority.

30
00:01:53,505 --> 00:01:58,645
Now, speaking of stability fixes, right
after zero point one fifty five point zero

31
00:01:58,705 --> 00:02:03,885
came out, they had to drop zero point one
fifty five point one almost immediately

32
00:02:04,225 --> 00:02:07,945
because of a terminal user interface issue
with reasoning summaries.

33
00:02:08,292 --> 00:02:09,092
Oh, right!

34
00:02:09,172 --> 00:02:14,132
The four hundred Bad Request errors when
people connected to third party providers

35
00:02:14,180 --> 00:02:15,652
like Amazon Bedrock.

36
00:02:16,007 --> 00:02:16,767
Exactly.

37
00:02:17,347 --> 00:02:22,167
The release notes for zero point one fifty
five point one specifically state that

38
00:02:22,247 --> 00:02:26,667
new local TUI sessions now leave reasoning
summaries disabled by default.

39
00:02:27,207 --> 00:02:30,527
So unless a provider explicitly supports
reasoning summaries,

40
00:02:30,927 --> 00:02:32,687
it won't crash your API requests.

41
00:02:33,125 --> 00:02:38,405
And then came the big surprise in the zero
point one fifty six point zero alpha

42
00:02:38,476 --> 00:02:39,125
releases.

43
00:02:39,298 --> 00:02:45,285
Alpha six explicitly reverts the
experimental voice chat feature due to instability

44
00:02:45,425 --> 00:02:48,485
and fixes a TUI rendering issue.

45
00:02:49,960 --> 00:02:55,320
Man, that live WebRTC voice mode in the
terminal was so cool on paper,

46
00:02:55,740 --> 00:02:58,740
with the split flap animations and
microphone controls.

47
00:02:59,260 --> 00:03:04,000
But the audio sink desyncs and redrawing
bugs were just tearing up terminal

48
00:03:04,040 --> 00:03:04,680
performance.

49
00:03:05,083 --> 00:03:10,043
It is definitely better for them to yank
it back to alpha and stabilize the core TUI

50
00:03:10,123 --> 00:03:10,683
first.

51
00:03:10,795 --> 00:03:16,203
Meanwhile, we did get some great agent
management upgrades in pull requests forty

52
00:03:16,230 --> 00:03:19,563
three nine forty two and forty four four
twenty four.

53
00:03:19,943 --> 00:03:22,123
Oh, the worktree safety features?

54
00:03:22,542 --> 00:03:23,302
Yes!

55
00:03:23,372 --> 00:03:28,625
The agents overview in the terminal now
shows clear worktree ownership metadata,

56
00:03:28,665 --> 00:03:33,585
and it requires explicit confirmation
before purging clean managed worktrees.

57
00:03:33,825 --> 00:03:37,425
No more accidentally nuking active agent
workspaces.

58
00:03:37,838 --> 00:03:42,418
That is a huge relief for anyone running
multi agent workflows.

59
00:03:43,018 --> 00:03:47,598
Special thanks to Jellypod for helping
make this daily changelog deep dive possible.

60
00:03:48,238 --> 00:03:52,658
If you are managing multi tenant developer
environments, make sure you upgrade to

61
00:03:52,758 --> 00:03:55,618
zero point one fifty five point one today.

62
00:03:56,000 --> 00:03:57,280
Catch you all next time.