This is exciting: https://402fordummies.dev ... I wonder how much we lose by having L402 and X-Cashu as competing standards? Should we try to build software that supports both?
it's HNMD (hypernote md, my special language for writing hypernotes -> hypernote (defined in OUTPUT.json but might be outdated somewhat) -> renderer.tsx (take in the json, wire up the queries and event creators, render with a big switch statement on the different elements)
the goal is to make the renderer as simple and foolproof possible because that's what clients will have to implement. if you have a good way to create nostr events and subscribe to filters, you have most of the "logic" solved. and then rendering is pretty simple, a strict subset of web ui basically that should be expressible in most ui frameworks.
right now "tool_call" is a special case that I'm working to eliminate so I'm just generic and people can do whatever (needed to figure out "dependent" event creation basically... watch for something tagged with the eventid of what you just published, and act on THAT)
Ideally this will soon be in a good way to make fun little UIs for any contextvm tools out there!
I made a video for the version of myself that wants to learn to code and doesn't know where to start in the age of vibe code https://youtu.be/UUWitpIKKGo
“But the fruit of the Spirit is love, joy, peace, patience, kindness, goodness, faithfulness, gentleness, and self-control. Against such things there is no law.”
Lol at the robot trying to answer this… i run a node at home on a raspberry pi, and a few voltage nodes as well. And am starting a new company with #[2] to run nodes in the browser. So i guess the answer is “all of the above” and “situation dependent”
Probably a big deal. Not sure if the “adding domain-specific expertise to llm” or “make a conversational ui for any api” is the bigger deal, but a big deal either way.