Curious about your reverse engineering efforts - hat was the trickiest part about getting Firebender’s background agents to run in fully isolated workspaces in Jetbrains IDEs? Did you run into issues with IntelliJ’s project management / sandboxes behind the scenes, especially when trying to keep the agent’s changes cleanly separated from your main code? How did you manage those edge cases or unexpected side effects

Really interesting launch, there’s lots like this to do in order to improve devex in the Jetbrains ecosystem

Building the "chrome tab" style experience for background agents was insanely tricky.

We have to open IDE project tabs without losing focus on the primary project which is impossible to do with the interface Jetbrains offers.

We reverse-engineered IntelliJ's project tab implementation and hacked our own version by calling internal IDE functions directly with reflection.

This was all to literally just open a tab in the IDE without losing focus, but it feels 10x better and is all worth it for our users.

The feedback loop between our users, mockups, and working react code is so tight with Onlook, it's fully replaced our Figma subscription

Question – is there a way to remove the selector ids from the generated code? We currently need to manually strip these from cursor

On the desktop app you'll be able to strip the IDs by stopping the run. There's no option for that yet on the beta web version. Let me know if that works :)

There are many disciplines (like Enterprise Architecture) that would benefit from being able to annotate entities and relations described here without extensive finetuning or running inference for these proprietary models; this is a huge unlock for research teams all over the world!

Mixture of Experts Model - https://en.wikipedia.org/wiki/Mixture_of_experts