Their seemingly copy/pasted generic response was basically "maybe negative usage balance, even though nothing told you about that, and also we can't actually check or confirm anything anyway".

@thsottiaux To add to the list of 'Usage tracking issues'; I hit my limit yesterday, was only a few hours till reset so not worth using a banked, so figured I would buy 250 credits to tide me over. Paid, never even saw them land, did no work, instantly gone. Support: "lol, 🤷"

I’ve now replied directly to @up_banking , and updated the same GitHub comment with it. Main constructive ask: if unsupported/private interfaces are restricted, pair that with a safer, capable, first-class Developer API; not just stronger prohibitions. github.com/up-banking/api/is…

Update: @up_banking has completed its internal review of my complaint. They acknowledged good-faith discussion was removed, apologised, and outlined plans to restore context + establish clearer guidelines/terms. Response + my initial thoughts: github.com/up-banking/api/is…

For the record: I’ve posted the formal complaint + this thread back to the GitHub issue, so subscribers should at least get it by email even if (likely when) @up_banking deletes it again. GitHub: github.com/up-banking/api/is… Wayback: web.archive.org/web/20260818… Archived for posterity.

Deleting discussion doesn’t make tooling disappear. It erases useful history, context and community knowledge, including criticism of those deletions. If this concerns you too, raise it with Up or share this thread so they see the impact on users who’d otherwise be advocates.

I’ve also raised a formal complaint directly with Up about how this was handled. The first half is attached here; continued below. My concern isn’t that Up has to endorse unofficial tooling. It’s the removal of relevant history/context, then criticism of those deletions too.

And this is getting increasingly disappointing. After Up deleted a large batch of relevant comments, I posted a comment criticising that approach. As expected, @up_banking has now deleted that too, apparently alongside yet more of the discussion. Screenshots for posterity:

Today I finally got around to buying Binary Ninja after wanting to for years, but never quite feeling the right time to justify purchasing. The 10 year anniversary + 35% off (10YRS35OFF) certainly helped motivate! These guys NEVER do discounts tbh. Keen! @vector35 @BinjaDevs

RE: xitter.cf/zats/status/2079298065…

It's extra useful when it divides your archived threads by project folder; high value semantics right there!

Feedback: Point Codex + GPT 5.6 Sol at the codex GitHub issues and properly triaging/cross-link/de-dupe/acknowledge and fix them; it honestly feels pretty shitty being a dev reporting things with proper investigation/etc just to have it rot in the void.

That feel when the Codex desktop app doesn't provide good window names, or good ability to search through / manage / persist open window state between updates/restarts/etc; so you end up reverse engineering ways to solve it yourself and build the supporting tooling to do so.

So uhh, @thsottiaux et al.. what do you consider to be an acceptable delay lag on @ChatGPTapp 's TTFT? Because personally/respectably.. 1min 10sec aint it... and it's been this way for the last week or two at least.

You can do this already with a scheduled prompt in the 'Automations' section. I believe it's even one of the default suggestions in the Codex app.

That is some mighty unreadable text colour / background choice combinations @todoist / @ChatGPTapp very un-accessibility friendly

And then extra bonus points if Codex CLI / Desktop app had similar tools that let me (optionally, with permission, when requested) give it access to call ChatGPT thread search / etc so I don't have to be the manual copy/paste glue between two contexts. cc // cc // @thsottiaux

I know I personally have access to search on ChatGPT with Cmd+K; but it would be nice if that same functionality could be exposed as a tool to ChatGPT itself, so that when I explicitly ask it to help me find an old thread, it can check multiple rather than me having to.

Also, naming the windows based on the thread they have open, so I don't have to manually sift through 13 different 'Codex' windows to try and and guess which one was the one I actually needed.

The wrapper reads the session ID printed at the end of a Codex session. Then the helper (which can also be run standalone) looks up the matching session log, extracts the token usage, and calculates the equivalent API cost from that.