Replies: 3 comments 1 reply
-
|
I'm very interested in this as well. For instance, I use claude code for coding sessions which I don't want to be stuffed with PAI context, so would love to be able to install it only in a specific project directory |
Beta Was this translation helpful? Give feedback.
-
|
Hey, just starting over with PAI and I was asking myself the same question, would be great to have some answer, thanks! |
Beta Was this translation helpful? Give feedback.
-
|
PAI is a global install — it lives in |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Thanks for putting PAI out there. I’ve been reading through the repo and docs and had a few questions I couldn’t quite answer for myself and I need some documentation before I try to run this (unless I put it inside an LXC, which i guess I could ... but I wanted to ask here anyway).
I’m trying to understand whether PAI is intended to be a global, always-on environment, or whether it can be cleanly contained to a single directory.
From what I can tell, everything is rooted at PAI_DIR (defaulting to ~/.claude), which suggests you could point it at a project-local .claude/ and only have it active when Claude Code is launched in that directory. But the docs mostly talk as if ~/.claude is the normal and expected setup, so I’m not sure if that’s just convention or an actual design assumption.
A few things I’m hoping you can clarify:
If I set PAI_DIR to a project-local path, is that enough to fully contain PAI there, or are there pieces (hooks, statusline, memory, etc.) that still assume a global scope?
Is “project-scoped PAI” something you consider a supported usage pattern, or more of an edge case?
Are there any packs or features that implicitly rely on ~/.claude or global state, even if PAI_DIR is set elsewhere?
Is there a reason the documentation centers so strongly on a global install? (e.g. assumption of a single long-lived assistant, shared memory across contexts, etc.)
I’m really interested in PAI’s workflow and reasoning structure, but I’m cautious about tools that feel like they want to become the default AI environment everywhere. If a localized/sandboxed mode is first-class (or could be), it would be great to see that called out explicitly in the docs.
Totally possible I’m just missing something obvious ... happy to be corrected. Thanks again for the work you’ve done on this.
Beta Was this translation helpful? Give feedback.
All reactions