Incidentally, Chroma also produced the single best study on long-context degradation that I've come across:

https://research.trychroma.com/context-rot

Before that, I cited nolima (https://www.reddit.com/r/LocalLLaMA/comments/1io3hn2/nolima_...) constantly to illustrate how difficult tasks involving reasoning or multi-step information gathering degraded much faster than the needle-in-haystack benchmarks cited by the major labs. Now Chroma is the first stop. Nice job on the research!

Cool project!

I would hope "native" means at least either CPU-native code (i.e. not Typescript) or OS-native UI (i.e. not Electron) - what did you mean by it here?

undefined

Very cool!