pxpipe started in September 2025 as a weekend experiment: rethinking how we should fill a model's context.
It was inspired by how humans went from the command line to the GUI. A GUI has far higher information density, and it can say that not all information deserves the same attention: what you focus on stays sharp, the rest recedes. LLMs haven't made that jump yet, we still feed them an append-only, stateless log of text, where every token gets the same weight. pxpipe renders that context instead and lets the model take it in through vision.
That's only a first step. The project drifted while I built it, from an interface question toward something closer to visual context compression, and I don't think it's the final shape of the idea.
I don't have the time to maintain all the issues and PRs here. I'll read what I can, but I can't promise responses or reviews.
That said, I'd encourage anyone interested to take this further than I have. The tradeoffs I know about are in the README.
Thank you all for the interest in the project.
pxpipestarted in September 2025 as a weekend experiment: rethinking how we should fill a model's context.It was inspired by how humans went from the command line to the GUI. A GUI has far higher information density, and it can say that not all information deserves the same attention: what you focus on stays sharp, the rest recedes. LLMs haven't made that jump yet, we still feed them an append-only, stateless log of text, where every token gets the same weight. pxpipe renders that context instead and lets the model take it in through vision.
That's only a first step. The project drifted while I built it, from an interface question toward something closer to visual context compression, and I don't think it's the final shape of the idea.
I don't have the time to maintain all the issues and PRs here. I'll read what I can, but I can't promise responses or reviews.
That said, I'd encourage anyone interested to take this further than I have. The tradeoffs I know about are in the README.
Thank you all for the interest in the project.