ChatGPT App is reading ~1.1Gb / sec all the time after upgrade. why??? anoter I/O bug in the latest version? this time it does not write, but reads A LOT!
@thsottiaux Reliable SOTA models (best for backend for sure), predictable spending and limits. Nice plugin system and integrations.
TODO:
1. Multi account support for single Codex App
2. Scroll buttons for tabs of sidebar (no way to scroll to hidden tabs now)
3. Luna grade model as subagent
@LLMJunky I did few attempts to create meshes of walking robots, but codex fails to do ortho projection right, although the concepts look nice. so far I tried to use https://t.co/ZAq3zFBrJD to make meshes, but it sucks so bad... IMO the ideal pipeline would be comfyUI + geom accurate model
@thsottiaux Fork is slow and expensive, could you guys borrow some ideas from git to make forks cheap and space efficient and glue threads from linked list session blobs?
@simpsoka 1. Multi account support, so that I can use same Codex App with personal and business accounts (e.g. account is picked per project and inherited for further spawned threads via codexapp tools).
2. Nested tag system. Currently large projects are messy and flooded with threads.
@thsottiaux 1. Multi account support, so that I can use same Codex App with personal and business accounts (e.g. account is picked per project and inherited for further spawned threads via codexapp tools).
2. Nested tag system. Currently large projects are messy and flooded with threads.
@thsottiaux@Dimillian Please add path like tags, so that agents could organize daily threads into more suitable structure (like virtual folders in gmail). Pinned threads are life saving, but they don't solve cross projects threads aggregation problem.
@thsottiaux Please add an ability to tag threads with pathlike tags, so that they can be seen in a virtual tree on the side bar (like gmail). This enables cross project grouping.
@ah20im May we have tags for threads so threads could be arranged not only by physical project, but could appear in virtual trees build of tags (like filters work in gmail). Simple, but very powerful feature where pinned threads could be just those with builtin "pinned" tag.
@thsottiaux@RhysSullivan Are there any plans for a new spark model? Kimi seems to fit 1T into Cerebras recently. To be honest context window size (128k) was a show stopper for gpt-5.3-codex-spark.
@romainhuet While planning complex goals, it would be nice to be able to clarify the codex suggested options as they often not provide enough context to fully understand codex intent. Currently have to Esc, or select "none of above" and put a massive note there to force codex explain options
@LLMJunky Ideas managed to reach 5% left mark, but 70% of agents were using fast mode though.
And by the way don't forget that you're using 2x limits until April 2nd. So it will be even worse.