Awesome. Are you exposing the api to Claude or just copy pasting the data for it to analyze? How are you feeding it the data?

It is extremely simple. I just tell Claude: "Use korb to look at my order history and create a table with my most frequently bought items. Next time, fill my basket with the items I always buy." So something of that variation works.

True. That's reporting and they are also reporting numbers internally, which are getting leaked.

Clean design. Would love to see trend lines over time.

really impressive for the size. Curious to see what happens when someone trains a 100B+ model natively at 1-bit.

The PyTorch export is key. Visual ML tools that lock you into their own runtime never stick. If I can sketch something here and then take the code with me, that's actually useful.

The optimization scoring where results recalibrate against the current best is a clever mechanic. Keeps it competitive even after the easy wins are taken.

The parallel agent coordination is what makes this interesting. Most agent wrappers are just single-agent loops with extra steps.

Yes, we actually added it because we were getting impatient with how long some implementations were taking. Appreciate this comment - thank you!