dropthejase makes Louis think out loud
A run of streaming-interface work that lets users watch the assistant reason in real time and cleans up the small visual glitches that came with it.
dropthejase has been polishing how the chat actually feels to use. The headline change: the assistant now streams its reasoning as it works, with a live "thinking" indicator and expandable blocks that show the model's chain of thought before it answers. For anyone using a legal assistant on a real matter, seeing the reasoning unfold builds trust in a way a bare answer never does.
The rest is the unglamorous craft that separates a demo from a product. Tool actions no longer flicker or pop back open as a response finishes, edit cards vanish cleanly once you accept or reject them, and a shared message view now powers both the main chat and project pages so they behave identically. There's also a small fix so short or unclear prompts get a sensible chat title instead of a cryptic one.
Spotted something wrong? Or know the PR text has fresher detail than the writeup above?