The goal was to bring down the cost at the context eng. level. We do it with Layout Memoization. Instead of dumping HTML into the context window, we have built a continual learning browser harness (read only for now). We have built an early prototype for you to try out, where you can: 1. Spins up a browser instance
2. Extract any structured or tabular data from anywhere on the open-web
3. And you can do all this at the cost of a vector search Would love to hear your thoughts on this. Thanks for ...