CMD+K connection failed since 0.49

Bug Reports
1

Hi,
It looks like I often get a connection failed issue with CMD+K.
I tried multiple models and it still didn’t work, but in Composer/Chat it worked.
Request ID: d781af09-e5a1-4201-bae2-eec5885063fb
Thanks!

2

Hi Hugo! It seems like you’re in privacy mode, so I’m unable to take a look at your request. If you’re comfortable doing it, would you be able to disable privacy mode, repeat the request, and send over the request ID? You can re-enable it after!

If not, would be great to get some console logs if you have any.

3

For sure, completely forgot about it. Next time it happens, I’ll disable it and make a new request for you to get a request ID I can share.

4

Alright, the issue happened and I disabled my Privacy mode to re-issue the request, here’s the ID: 3e25d79d-5c25-47e0-9653-8364de81e46c (Mac) and d88784ed-a47f-4cf3-a882-0071bc3f1405 (Windows)

Let me know if you can access it correctly now.

Thanks!

5

If you disable a few rules completely, does it reproduce? I wonder if it’s because you might have some rules that are too big for CMD+K!

(Nevertheless, we should just trim them rather than block you on execution)

6

That could be it! My rules are minimally 13K tokens long, but the prompt in the CMD+K when the issue happened was not long, it was like 1-2 sentences, not more than that.

7

Interestingly when the issue happens, I can still use Haiku, but DeepSeek (and other DeepSeek models) and 4o-mini (and other OpenAI models) are not working.

8

@dcastl The issue keeps happening and when it happens only Anthropic models work for some reason. Any ways I can fix this issue or help fix it (happy to hop on a call and help debug)?
Because, not being able to use CMD+K 90% of the time is actually quite irritating…
Also, interestingly, it’s only the “generate edit” that doesn’t work, if I use the “quick question” option, it works.
Thanks!

Request IDs:

  • 3e33b094-3c7d-4ffc-8361-3080b8979216 (gemini)
  • 7dea5004-9d18-4bb0-b075-e3c269a7a860 (deekseek)
  • 07e105fa-c539-47c4-b267-b5408c908079 (grok)
  • 4ac49a89-cb1f-4b48-a0c8-a33147ea0994 (openai)
9

I think the issue is due to the rules length. I see some really big rules in these requests (like AI Memory System) that also get applied to CMD+K, causing it to go over the token limit!

10

What confuses me is that the issue only happens with CMD+K and not Composer, while they both have more or less the same rules. Maybe you could help me understand how they both behave?

PS: I reviewed my rules (I had some duplicated ones applying at the same time), and now it works fine, but it would still be smart in the long run to either show a message saying the rules are too long or are you said “just trim them rather than block you on execution”.