Conflicting home directories .config/PAI-Pi and .pi/agent

[henkwerner]

The Pi INSTALL.md (at https://github.com/danielmiessler/Personal_AI_Infrastructure/blob/main/Releases/Pi/INSTALL.md) states that the scaffold has home directory .config/PAI-Pi. This is also reflected in other documents, such as the extensions/pai-core/index.ts.

However:
The install command npm install -g @mariozechner/pi-coding-agent (which leads to cat "/opt/homebrew/lib/node_modules/@mariozechner/pi-coding-agent/README.md") sets .pi/agent as the home directory.

Because of this, Pi is unable to find any of the system files. I had to move the models.json to .pi/agent to enable llama-server to find it.
Pi has no identity until it reads the system files at launch.

So, which one is it?

1. Should I move all my files from .config/PAI-Pi to .pi/agent? This would mean I cannot use any template from the repository without editing every mention of the home directory.
2. Or should I reprogram my Pi to look at .config/PAI-Pi at loading?

[mpsix]

Same issue here. Took me awhile to get past this step but finally figured out to use the following environment variable PI_CODING_AGENT_DIR

I set it as PI_CODING_AGENT_DIR=~/.config/PAI-pi and it picks up PAI-Pi now but I'm still having issues with it connecting to Ollama. The models.json file format described doesn't align to what Pi wants to see (fixable changing the models section to an array). Fixed that but still having issues. Hope this helps you on the right path.

[henkwerner]

Thanks @mpsix , I will have a try on it.

For reference, here is my models.json for llama.cpp and bonsai. Perhaps you can edit it for your Ollama setup.

`{
"providers": {
"llama-cpp": {
"type": "llama-cpp",
"baseUrl": "http://localhost:8080/v1",
"api": "openai-completions",
"apiKey": "sk-no-key-required",
"models": [
{
"id": "Qwen3.5-9B-Q4_K_M.gguf",
"name": "Qwen3.5 9B (Local)",
"reasoning": true
}
]
},

"bonsai": {
  "type": "openai",
  "baseUrl": "http://localhost:8081/v1",
  "api": "openai-completions",
  "apiKey": "sk-no-key-required",
  "models": [
    {
      "id": "/Users/jeannie/Bonsai-demo/models/Bonsai-8B-mlx",
      "name": "Bonsai 8B (Local)",
      "reasoning": true
    }
  ]
}

}
}`

[aplawson7707]

I'm still testing this locally, but I saw here that there was a minimal Ollama example that might also be helpful along with the llama-cpp config posted by @henkwerner

{
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434/v1",
      "api": "openai-completions",
      "apiKey": "ollama",
      "models": [
        { "id": "llama3.1:8b" },
        { "id": "qwen2.5-coder:7b" }
      ]
    }
  }
}

[aplawson7707]

Same issue here. Took me awhile to get past this step but finally figured out to use the following environment variable PI_CODING_AGENT_DIR

I set it as PI_CODING_AGENT_DIR=~/.config/PAI-pi and it picks up PAI-Pi now but I'm still having issues with it connecting to Ollama. The models.json file format described doesn't align to what Pi wants to see (fixable changing the models section to an array). Fixed that but still having issues. Hope this helps you on the right path.

@mpsix this might be a dumb question, but where does this .env file need to live?