Welcome to Discussions About the PAI Project!

[danielmiessler]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.

To get started, comment below with an introduction of yourself and tell us about what you do with this community.

[smolcompute]

Oh! Awesome! This will be super useful. I've just implemented the UFC last night using Codex CLI instead of Claude and it seems to be working!

One question / thought I have: I wanted PAI to be model agnostic so built it inside a ~/. pai folder so that there are no references to either a ~/. claude or ~/. codex folder. All the paths the AI is instructed to hydrate from are in ~/. pai. But would it be better to avoir any absolute paths in the UFC, only use relative paths so that the file structure can be copied and dropped inside any model's folder and it works without having to update any paths ?

[danielmiessler]

Really great idea. Let me add that. Because the platform is supposed to be independent from provider anyway, so this is a really good point.

[smolcompute]

I saw the new PAI_DIR variable which is great. Could it be possible / a good idea to also replace the .claude folder by something else in the Github repo ? Right now I think users have to keep it named .claude in order to pull updates from Github. Maybe just name it pai or .pai ? And people can create symlinks of this folder, rename them to .claude, .codex etc., and place them in their home directories ? Would this work?

[smolcompute]

Thanks for the update ! Love to see PAI becoming agent-agnostic. Now we just have to wait for Anthropic to adopt AGENTS.md 😅. Symlinks in the meantime.

[muhammedshammashere]

👋 Hi everyone!

I’m Muhammed Shammas. I'm a Web Developer and enjoy playing around with whatever is fresh in tech and sharing what i find.#Tech. I'm here to meet the community, learn from others, and provide value where I can.

Here are few of my projects on Github: https://github.com/muhammedshammashere 💪

[Geelhem]

I am Guilhem, teacher and looking at future proofing my note taking for future ai enhancement. Could this be used with ollama? Or open web ui via mcp? Love all the ideas about giving structure to personal knowledge.

[danielmiessler]

Not yet but I'm working on those. Daniel

…

On Fri, Sep 26, 2025 at 3:25 PM, Geelhem < ***@***.*** > wrote: IAM Guilhem, teacher and looking at future proofing my note taking for future ai enhancement. Could this be used with ollama? Or open web ui via mcp? — Reply to this email directly, view it on GitHub ( #11 (comment) ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAAMLXR3TBOR7KEZHLKIYBT3UW4N3AVCNFSM6AAAAACHBU5E2GVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTINJSGY2TKNA ). You are receiving this because you authored the thread. Message ID: <danielmiessler/PAI/repo-discussions/11/comments/14526554 @ github. com>

[danielmiessler]

Oh great idea

…

On Sat, Sep 27, 2025 at 13:35, smol compute < ***@***.*** > wrote: I saw the new PAI_DIR variable which is great. Could it be possible / a good idea to also replace the .claude folder by something else in the Github repo ? Right now I think users have to keep it named .claude in order to pull updates from Github. Maybe just name it pai or .pai ? And people can create symlinks of this folder, rename them to .claude, .codex etc., and place them in their home directories ? Would this work? — Reply to this email directly, view it on GitHub ( #11 (reply in thread) ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAAMLXXPFACSELDRCSIH6GL3U3YJDAVCNFSM6AAAAACHBU5E2GVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTINJTGEYDEMA ). You are receiving this because you authored the thread. Message ID: <danielmiessler/PAI/repo-discussions/11/comments/14531020 @ github. com>

[smolcompute]

I think it's better if the repo's .claude folder is renamed to something that is not a dot file, because the hidden folder causes some interoperability issues with other apps. For example Obsidian refuses to display these folders, or even non dot file symlinks pointing to them. If I open the parent PAI folder as a vault, none of the markdown files inside .claude show up. So I'd argue in favor of just naming it pai or pai_dir or whatever and renaming the symlinks pointing to it .claude, .codex etc.

[danielmiessler]

100

…

On Sun, Sep 28, 2025 at 16:46, smol compute < ***@***.*** > wrote: Thanks for the update ! Love to see PAI becoming agent-agnostic. Now we just have to wait for Anthropic to adopt AGENTS.md 😅. Symlinks in the meantime. — Reply to this email directly, view it on GitHub ( #11 (reply in thread) ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAAMLXRQZ2PCAFQGHSNQNOD3VBXM7AVCNFSM6AAAAACHBU5E2GVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTINJTGY2DOOI ). You are receiving this because you authored the thread. Message ID: <danielmiessler/PAI/repo-discussions/11/comments/14536479 @ github. com>

[jdax57]

Excuse my ignorance, but I see that we call PAI vendor agnostic, but how is this merged or leveraged with other CLI terminals. Such as Codex, Gemini, or Opencode. Seems that this heavily rests on claude code. Again this may be my ignorance or lack of knowledge as a noob. Eitherway can there be a section added on how to enforce or setup agnostic vendors. Even such as calling other terminal CLI assistants. NetworkChuck showed using CC to call Gemini to offload work. Which is brilliant when you want to off load work to your local AI infrastructure vs the cloud.

[danielmiessler]

You just have the Gemini command in your prompt. Check the research skill. We have it already!

…

On Wed, Oct 29, 2025 at 20:54, jdax57 < ***@***.*** > wrote: Excuse my ignorance, but I see that we call PAI vendor agnostic, but how is this merged or leveraged with other CLI terminals. Such as Codex, Gemini, or Opencode. Seems that this heavily rests on claude code. Again this may be my ignorance or lack of knowledge as a noob. Eitherway can there be a section added on how to enforce or setup agnostic vendors. Even such as calling other terminal CLI assistants. NetworkChuck showed using CC to call Gemini to offload work. Which is brilliant when you want to off load work to your local AI infrastructure vs the cloud. — Reply to this email directly, view it on GitHub ( #11 (comment) ) , or unsubscribe ( https://github.com/notifications/unsubscribe-auth/AAAMLXX2OJ3VE6BUAIA4STL32ELNZAVCNFSM6AAAAACHBU5E2GVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTIOBRHE3TANY ). You are receiving this because you authored the thread. Message ID: <danielmiessler/Personal_AI_Infrastructure/repo-discussions/11/comments/14819707 @ github. com>

[appressman]

In my quest for overall utility of PAI I'm hosting it on an always on server. That way it can be doing things like checking mail, research etc always. I'd like to still use it with work environments like my laptop and phone. I was wondering what solutions anyone might have with similar needs. My quick research leads to a mix of Claude Desktop and a web UI on the PAI server with details in the attached. I'm open to revision/replacement of this approach from those farther along than me here.
Accessing Your Personal AI from Windows and iPhone.docx

[jdax57]

I've been running it on a WSL instance, after some pitching around and modding it (Voice server leans heavy into Mac UNIX). I managed to use it with Gemini, I was going to test it with codex (however seems codex is still new), I was also planning to test it with opensource. My research sadly leads me to state that Claude CLI is sitll a tad ahead of the power curve from others. As for access via phone, I'd suggest N8n, a DNS for access, and webhooking to n8n. That is sadly above my knowledge level.

Sorry I did not open your docx....I sadly don't trust word docs....Maybe post a png or an md file?

[appressman]

Yeah, sorry myself about not thinking about .docx files. Here's a .md
Windows - iPhoneAccessToPAI.md Ultimately being able to "Grok" my PAI on my phone would be grail I seek.

[jdax57]

Very abitious. while I'm still learning this, I would suggest tailscale and avoid posting up to web, keep it all home and private network centric. But what do I know?

[appressman]

Thanks. All Tailscaled.

…

On Nov 7, 2025 at 5:31 PM -0500, jdax57 ***@***.***>, wrote: Very abitious. while I'm still learning this, I would suggest tailscale and avoid posting up to web, keep it all home and private network centric. But what do I know? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: ***@***.***>

[GaryGealy]

Hi I completed the setup instructions, but when I start Claude code and type

"Hey, tell me about yourself"

I get the following. I suppose this means the installation is incorrect? The docs don't say what exactly I should be getting in response.

⏺ I'm Claude Code, an AI-powered CLI tool built by Anthropic to help with
software engineering tasks...

Not even sure if this is the right place to ask this sort of Q, please direct me to the correct place if its not.

I have been using Claude Code for over a month before trying the PAI install. Not sure if that could be an issue.

Thanks

[sudo-eugene]

@GaryGealy you're not the only one #95

[bcrawford098]

I stumbled upon Network Chuck's video about TELOS and then found Claude code. I didn't start with PAI. I began by creating personal context files to establish a central context repository that defines how I want to:

• Work with project folders (Git, directory structure, documentation, tools, etc.).
• Manage context and schema files specifically for my local Obsidian vault.
• Handle efficiency and project management.
  • This includes tracking STATUS.md, creating session summaries, updating my daily note, committing/pushing, etc.

Anyway, I soon discovered PAI. Instead of just setting it up from scratch and importing my context files, well, let's just say it was a reminder to step back and analyze a problem from more than one angle.

Currently, I have a meta-project in a folder named for my personal assistant. I use this project to add functionality to my PAI. Essentially, I instruct it, "Using the PAI framework and best practices, I'd like to set you up to work with my Obsidian vault."

Since I had already configured my research assistant portion to utilize Claude, Gemini, and Perplexity, it was able to research the best Master Control Program (MCP) method for my setup and needs. (I ended up using a hybrid approach, as most of my projects reside within my vault.)

I can also discuss and interact with other projects, which is great because if I change something globally, I can see how it might affect my other projects. Although, I have been deliberate in working towards not needing to do this. My cleanup is mainly a result of how I migrated from my previous workflow to using PAI.

Much like Obsidian itself, the PAI infrastructure is a highly extensible framework designed to optimize working with numerous tools. This can feel overwhelming because the possibilities and approaches you might take are up to you. I've found that the more specific and detailed I am in explaining to my PAI how I want my projects and tools to function, the better the results. I've only been working with PAI for a week, and I've already created a roadmap of functionality I want to build for myself.

Finally, I'd like to mention that discovering the TELOS framework has been both timely and revolutionary for me. Finding and being able to use PAI to plan and execute my strategies, goals, and projects feels almost like kismet and may have just saved my livelihood.