idk what Eliza is, this is my own WIP thing which consumes all the data you give it and classifies it as needs human input vs.

idk what Eliza is, this is my own WIP thing which consumes all the data you give it and classifies it as needs human input vs. ignore, then if the former, classify into a category, then that is fed to a (relatively) more advanced LLM which generates the chip and decision actions to click on (and notifies you) The last part is cheap, because it's the equivalent of <100 Luna/Sonnet calls a day, trivial cost. The expensive part is sifting through unfathomable amounts of garbage data of indeterminate format. Emails and chat messages aside it consumes raw sensor outputs from IoT/smart home devices, health/fitness trackers that have different APIs, location trackers for belongings/pets, banking, going through massive camera rolls There are no standard APIs that cover all possible consumer data stream combinations and so this must be LLM-routed, which was always the most expensive part (with clear intelligence/cost tradeoffs) - until now