They have a flag Zero Data Retention that cones with the promise that after a call no data is kept on any servers. I’m using that.
It’s contract level trust; I wouldn’t bet my life on it, but overall it’s a compromise I’m willing to make.
I run the AI service (OpenwebUi) myself, but buy the compute on OpenRouter. Self hosting models is not an economic viability for me.
For how to use bash, you can easily resort to local models with LMStudio. No internet search needed.
Drop chats, use a harness. Include somewhere the procedure instructions to always look up documentation before crafting the command. Add instructions somewhere what dependencies you tolerate. Then it’s really good.
Once your tasks get larger, you need additional checks.
Provide the URL to your AI chat and ask it to print out the article for you.
Pi Coding Agent on the laptop, openwebui as a self hosted service.
I don’t host LLMs myself. My best hardware has 16GB of RAM, I could run a 4B model or so, but what’s the point? So I buy compute with ZDR turned on.