Model capacity is solved 🥀

Dots are really good!!

Really hate this behaviour, its like Codex will explicitly not compact so it can finish its turn meaning your next message will auto-compact

We are live!

A very nice quality of life improvement is being deployed out right now to the @raygunio MCP server I've been wanting to add OAuth for a while so it would be even easier to set up the Raygun MCP server and now that's possible

Been using the new v3.0 APM profiler from @raygunio in @autohiveai to improve the performance of our virtual filesystem for agents Having such a lightweight profiler now is huge as we can run it in production and start getting agents looping on performance improvements

Just saw this thumbnail so I had to

I really want to love GitHub's stacked PRs but I just can't get them working. Tried twice now with two different streams of work and they just get stuck in merging forever without any info on what is blocking it

Trying out Dia again for a bit, hopefully it sticks this time as there is a lot of nice design changes compared to Arc but there's also a lot I still love that Arc has and Dia doesn't

Spaces are in @diabrowser now!! This has been one of the bigger things I've been wanting before I moved off Arc, multiple windows for profiles was a hassle to manage, would love to see icons/emojis as well! Only think left now is the Command Bar without going to a new page ;)

Just found out that Buzz resends instructions on how to use Buzz and the conversation history on every user message, that seems really wasteful for context and caching

Running into some weirdness with Kimi K3 on @FireworksAI_HQ when using the Responses API. It will sometimes start outputting Python repr looking objects as the visible text and not actually do the tool calls it's saying it will do causing premature stopping of agents This doesn't appear to happen on previous Kimi models and it doesn't happen on the Chat Completions API. It only appears to happen on the Responses API Example below