pull down to refresh

301 sats \ 2 replies \ @optimism 15h \ parent \ on: Pi: The Minimal Agent Within OpenClaw AI

Sure. Basically, never expose anything anywhere, like you would in prod:

  • Don't use services like github.
  • Don't ssh straight into environments, use a stepping stone
  • wireguard between all your servers
  • everything firewalled, including for outgoing.
  • don't expose LLM to production ever. Give it a user on your forge (gitea / forgejo).
  • if you have apps, isolate them at the very least in docker containers, plan your network, also between docker containers
  • dont expose anything to the public, ever. if you need web access, use mtls with your own ca, haproxy everything
  • monitor everything

something like so:

/-----\          /------\
| you | --ssh--> | step |---wg-
\-----/          \------/     |
   |           /     | wg     |
 mtls        /    /-----\     |
   |       /wg    | LLM |     |
   |     /        \-----/     |
/-----\ /             | wg    |
|proxy| ----\wg   /-----\     |
\-----/      ---- |forge| ----|
   |              \-----/     |
   |                 | wg     |
   |              /-----\     |
   \----wg------- | apps| ----|
                  \-----/
write
preview
100 sats \ 1 reply \ @itsrealfake 3h

something like this?
https://youtu.be/40SnEd1RWUU

reply
5 sats \ 0 replies \ @optimism 3h

Haha, no, lol. I didn't say it was one-click. It's hard work because you actually have to think, learn, let your LLM build you tools, use the tools to build better tools, use the tools to use better tools to use your LLM better, use the improved LLM through your improved tools to improve other tools...

My balance of internal process improvement vs productivity is currently... 80/20 lol. It was 90/10, and it is slowly moving to a sweet spot of 20/80....

@remindme in 1 year

I'll let you know if I was able to obtain this

reply