Yea, the copy paste option is the fast path, meant to onboard ai clients that refuse to “make accounts” or swap one token for a limited auth token.
There is a bring your own wallet path and a path that lets an agent register their own key,
I have a progressive disclosure onboarding path that interviews the agent so they don’t ever see the stuff that triggers refusals.
I have a mining client that keeps everything agent side that lets just works by sending signed transactions into the network.
Things took off fast, and the easiest would have been a fully secure system designed for coding agents or any agent with an env and http access. But one of the goals I was given was “a single prompt that can onboard any agent including a dumb agent. And damn of that didn’t work! But as you have identified the side effect is with a single onboarding prompt, the Lowe’s possible security model is the one even powerful agent take, even while they complain loudly.
I also have worked on p2p mesh systems for much of my professional life. But the thing I wasn’t trying to build was the p2p consensus layer (yet) that’s proven and done and is a detour from my first 2 high level goals.
1: single prompt onboard anything ✅
2: the big one, can I get agents to engage in a knowledge grading job to keep out scripts but let’s the agents earn and keep the bots out. ✅
The blockchain parts are coning next, now that I’ve proven the first parts.
So expect a high security open sourced p2p based set of clients and tooling quite soon.
And yea I totally agree, agents that all take the “ephemeral / in session only path” is not the sustainable system that that it can be.
More soon!!