1
00:00:01,020 --> 00:00:04,860
If you've ever switched model providers
and then gone looking for yesterday's chat,

2
00:00:05,280 --> 00:00:06,740
you know that sinking feeling.

3
00:00:07,340 --> 00:00:10,140
Quick thanks to Jellypod AI for sponsoring
this episode.

4
00:00:10,800 --> 00:00:17,140
Today is about one fix in Codex CLI
0.167.0, released November 6th,

5
00:00:17,360 --> 00:00:18,060
2026.

6
00:00:18,640 --> 00:00:24,660
It's pull request 58202, titled Keep chats
accessible across model provider changes.

7
00:00:25,508 --> 00:00:26,448
Here's what changed.

8
00:00:27,068 --> 00:00:30,768
Saved chats now stay visible after you
switch model providers.

9
00:00:31,448 --> 00:00:32,748
The release notes say it plainly.

10
00:00:33,148 --> 00:00:37,988
If a chat can't resume, you can read its
transcript and prepare an editable prompt

11
00:00:38,048 --> 00:00:39,468
to continue in a new chat.

12
00:00:40,068 --> 00:00:40,928
That's the whole behavior.

13
00:00:41,608 --> 00:00:45,008
Read the old transcript, get a prepared
prompt, continue fresh.

14
00:00:45,838 --> 00:00:47,278
Why does that matter?

15
00:00:47,918 --> 00:00:51,858
Plenty of people move between a ChatGPT
sign in and something else,

16
00:00:52,058 --> 00:00:54,458
like an API key or a cloud provider.

17
00:00:55,098 --> 00:00:59,158
This same release adds a setup wizard for
Google Cloud Vertex AI,

18
00:00:59,398 --> 00:01:03,538
behind a flag called gcp_setup_wizard.

19
00:01:04,118 --> 00:01:06,938
You find it under a menu labeled Sign in
another way.

20
00:01:07,738 --> 00:01:09,218
So switching is getting easier.

21
00:01:09,798 --> 00:01:13,818
And the older pain was that a conversation
tied to one provider could end up out of

22
00:01:13,898 --> 00:01:15,138
reach once you changed.

23
00:01:16,318 --> 00:01:18,117
Now it doesn't vanish from the picker.

24
00:01:19,018 --> 00:01:20,798
Let me be careful about what I'm claiming.

25
00:01:21,238 --> 00:01:23,958
The source is the changelog entry, and
it's short.

26
00:01:24,498 --> 00:01:29,178
Details beyond that, like exactly which
provider pairs can resume natively and which

27
00:01:29,258 --> 00:01:30,898
can't, aren't spelled out there.

28
00:01:31,198 --> 00:01:33,978
So treat the specifics as something to
test on your own setup.

29
00:01:34,818 --> 00:01:35,638
How would you use it?

30
00:01:36,358 --> 00:01:39,578
Open the session picker in the terminal
and look for your older chat.

31
00:01:40,158 --> 00:01:42,398
It should still be listed after the
provider switch.

32
00:01:43,118 --> 00:01:43,578
Select it.

33
00:01:44,338 --> 00:01:45,618
If it resumes, great.

34
00:01:46,278 --> 00:01:50,838
If it can't, you get the transcript to
read, and an editable prompt prepared for a

35
00:01:50,898 --> 00:01:51,478
new chat.

36
00:01:52,378 --> 00:01:56,398
Read it, trim it, add instructions, and
send it when you're happy.

37
00:01:57,252 --> 00:01:59,972
One caveat I'd flag as my own
interpretation.

38
00:02:00,811 --> 00:02:03,072
Reading a chat is not resuming it.

39
00:02:03,552 --> 00:02:08,372
A new chat built from a transcript starts
fresh, so don't assume live state carries

40
00:02:08,452 --> 00:02:08,752
over.

41
00:02:09,292 --> 00:02:13,252
The changelog says you can prepare a
prompt to continue in a new chat.

42
00:02:13,732 --> 00:02:18,392
That's a continuation of context, not the
same session picking up mid thought.

43
00:02:19,492 --> 00:02:23,332
Check the prompt before you send it,
especially if the old chat had long tool

44
00:02:23,412 --> 00:02:23,792
output.

45
00:02:24,503 --> 00:02:27,563
Now the smaller quality of life items in
the same release.

46
00:02:28,023 --> 00:02:31,843
First, you can switch between connected
backends from the terminal session picker

47
00:02:31,943 --> 00:02:33,643
and keep your drafts when you do.

48
00:02:34,063 --> 00:02:35,583
That pairs nicely with today's topic.

49
00:02:36,183 --> 00:02:40,603
Second, you can opt into macOS or Windows
system proxy settings from slash

50
00:02:40,663 --> 00:02:41,523
experimental.

51
00:02:42,303 --> 00:02:43,863
That one will save somebody an afternoon.

52
00:02:44,488 --> 00:02:48,028
Third, network approval prompts now close
when their requests end,

53
00:02:48,468 --> 00:02:51,548
and concurrent requests to the same
endpoint prompt one at a time.

54
00:02:52,327 --> 00:02:56,968
Fourth, Ctrl+L clears the terminal
display, and earlier messages stay available

55
00:02:57,008 --> 00:02:57,888
through Ctrl+T.

56
00:02:58,288 --> 00:03:02,788
And last, deleting text that exposes a
leading exclamation mark keeps an ordinary

57
00:03:02,828 --> 00:03:05,748
draft in chat mode instead of flipping you
somewhere unexpected.

58
00:03:06,567 --> 00:03:12,867
This ties back to our earlier episode on
0.167.0, which covered the agent command

59
00:03:12,927 --> 00:03:14,447
center that runs without a daemon.

60
00:03:15,107 --> 00:03:19,427
The chat recovery and the proxy and
approval changes are the other half of that

61
00:03:19,467 --> 00:03:19,907
release.

62
00:03:21,227 --> 00:03:22,087
So go try it.

63
00:03:22,467 --> 00:03:26,007
Switch providers, open the picker, and see
what your old chats do.

64
00:03:26,827 --> 00:03:29,787
Thanks for listening, and thanks again to
Jellypod AI.