I don't really run any "agents" like that live at the moment but always willing to `docker-compose up` to test something.

Been loving GLM 5.2, finally works right in Cursor... haven't used it directly as an API yet though but I suspect that'll be more expensive than through Cursor Pro allotment. 

Grok 4.5 being 50% off in Cursor right now derailed using it a bit this week, it feels like what I expect Composer 3 should be... cheap, relatively fast, and good enough.

> Just publish an llms.txt on your domain

Experimenting with this for nodestr, will hit you up to try it... on-demand VPS.

justin_shocknet

With just GLM-5.2 (regular, not the expensive "fast" one) on my production work, I mostly spend a bit under $1 for each hour it is plowing away at things, single-threaded because I can't review that fast anyway.

Currently it's been on the lower end - about $2 a day - because I'm hitting some bugs in codex and reading logs through sqlite is a nasty chore, so I switch between pi+ppq and claude to compensate and get shit done at the moment (though since I downgraded Claude to the $20 sub, it basically means 3 prompts and its done for the next 5h, lol.)

Started looking at building on Pi a little, but on the backburner... I ultimately want people to be able to use the full shocknet stack through a chatbot...

Just publish an llms.txt on your domain and describe everything in markdown. Like a remote skill file. That should work for most implementations

What's the best Replit-like tool?

Undisciplined

> How much are you spending through PPQ?

With just GLM-5.2 (regular, not the expensive "fast" one) on my production work, I mostly spend a bit under $1 for each hour it is plowing away at things, single-threaded because I can't review that fast anyway. 

Currently it's been on the lower end - about $2 a day - because I'm hitting some bugs in codex and reading logs through sqlite is a nasty chore, so I switch between pi+ppq and claude to compensate and get shit done at the moment (though since I downgraded Claude to the $20 sub, it basically means 3 prompts and its done for the next 5h, lol.)

> Started looking at building on Pi a little, but on the backburner... I ultimately want people to be able to use the full shocknet stack through a chatbot...

Just publish an llms.txt on your domain and describe everything in markdown. Like a remote skill file. That should work for most implementations