Makes absolutely sense. I would do the same. And for now I only trust Goose as a AI agent on my daily driver, but it’s not running in an autonomous mode.

You should be paranoid.

Brother... Im scared you haven't already XD

i use an old macbook running linux for that. it’s free

I did this, then ended up giving it my keys anyway to do deployments 👩‍💻

running my code env on a remote server, for long running agent specifically, not by paranoia. But yes it makes sense

Sandboxing is good practice in these days.

Why not host that machine yourself. Like on a Proxmox or so. Gives you multiple envs if needed. Another thing I noticed is that dev resource consumption (CPU and mem) goes up significantly (Rust in my case) when multiple agents work, test and use tools (LSP) simultaneously. So 16 core 64gb does not ...

Are you running OpenClaw on your primary dev machine?

You are not being paranoid, it is a real problem. Using nostr:nprofile1qqs8msutuusu385l6wpdzf2473d2zlh750yfayfseqwryr6mfazqvmgpy4mhxue69uhkvet9v3ejumn0wd68ytnzv9hxgtm0d4hxjh6lwejkuar4wfjhxqfswaehxw309a5hgcmg0ykkwmmvv3jkuun0vskkvatjvdhkuargdacxsct8w4ejuumrv9exzc3wd9kj7qfpwaehxw309ahx7um5wgkhyetvv9uju...

What are they doing? Like what’s the goal?

Agents don't belong anywhere near sensitive data. Own machine/server/vps or at least container.