nixos/henrovnix_ok/home/apps/ai.nix

{ config, pkgs, ... }:

{
  # Install ZED and Ollama (Vulkan for CPU/AMD, use `ollama` for NVIDIA or `ollama-rocm` for AMD ROCm)
  home.packages = [
    pkgs.ollama-vulkan  # For Vulkan (CPU/AMD). For NVIDIA: pkgs.ollama. For AMD ROCm: pkgs.ollama-rocm
    pkgs.zed
  ];

  # Environment variables for ZED and Ollama
  home.sessionVariables = {
    OLLAMA_HOST = "http://127.0.0.1:11434";
    MISTRAL_API_KEY = "CWo91GHwIClzLj6bCLQ69IioSi54PpTZ"; # Replace with your actual Mistral API key
  };

  # Configure Ollama as a user service (starts with login)
  home.services.ollama = {
    enable = true;
    package = pkgs.ollama-vulkan;
    onStart = ''
      # Start Ollama server
      ${pkgs.ollama-vulkan}/bin/ollama serve > /dev/null 2>&1 &
      sleep 5  # Wait for server to start

      # Pull coding and chat models at startup
      ${pkgs.ollama-vulkan}/bin/ollama pull codellama:70b    # Best for coding
      ${pkgs.ollama-vulkan}/bin/ollama pull mixtral:8x7b    # Best for chat

      # To pull additional models, uncomment or add lines below:
      # ${pkgs.ollama-vulkan}/bin/ollama pull llama3:8b       # General-purpose
      # ${pkgs.ollama-vulkan}/bin/ollama pull qwen2.5-coder:7b # Multilingual coding
      # ${pkgs.ollama-vulkan}/bin/ollama pull qwen2.5-coder:32b # Larger coding model
      # ${pkgs.ollama-vulkan}/bin/ollama pull starcoder2:15b  # Alternative for code
    '';
  };

  # Configure ZED to use Ollama and Mistral API
  home.file.".config/zed/settings.json".text = lib.mkForce ''
    {
      "mistral": {
        "apiKey": "$MISTRAL_API_KEY",  # Uses the environment variable set above
        "defaultModel": "mistral-pro"  # Default model for Mistral API calls
      },
      "ollama": {
        "endpoint": "$OLLAMA_HOST",    # Connects to local Ollama instance
        "defaultModel": "codellama:70b" # Default model for Ollama plugin
      },
      # Add other ZED plugin configurations here if needed
    }
  '';

  # --- Notes ---
  # 1. Pulling Additional Models:
  #    To pull more models later, run:
  #      ollama pull <model-name>
  #    Example: ollama pull llama3:8b

  # 2. Switching GPU Backends:
  #    - For NVIDIA: Replace `ollama-vulkan` with `ollama` (uses CUDA)
  #    - For AMD: Use `ollama-rocm` and ensure ROCm is installed

  # 3. ZED Plugin Setup:
  #    - Install the Ollama and Mistral plugins in ZED via the plugin marketplace
  #    - The Ollama plugin will use the models pulled above
  #    - The Mistral plugin will use the MISTRAL_API_KEY for cloud access

  # 4. Custom Prompts:
  #    To add custom prompts for Ollama, create a prompts.json file or
  #    configure prompts directly in the ZED Ollama plugin settings

  # 5. Resource Management:
  #    Ollama runs as a user service and stops when you log out
  #    To run Ollama persistently, consider a systemd user service with `systemctl --user`
}

{ config, lib, pkgs, ... }:
let
  # Continue gebruikt tegenwoordig bij voorkeur config.yaml; config.json bestaat nog
  # maar is “deprecated” in de docs. We schrijven hier bewust config.json omdat jij dat vroeg.
  continueConfigJson = builtins.toJSON {
    models = [
      {
        title = "Qwen2.5-Coder 7B";
        provider = "ollama";
        model = "qwen2.5-coder:7b";
        apiBase = "http://localhost:11434";
      }
      {
        title = "Qwen2.5-Coder 32B";
        provider = "ollama";
        model = "qwen2.5-coder:32b";
        apiBase = "http://localhost:11434";
      }
      {
        title = "StarCoder2 15B";
        provider = "ollama";
        model = "starcoder2:15b";
        apiBase = "http://localhost:11434";
      }
    ];
    # Tab-autocomplete model (pas aan naar smaak/VRAM)
    tabAutocompleteModel = {
      title = "Qwen2.5-Coder 7B";
      provider = "ollama";
      model = "qwen2.5-coder:7b";
      apiBase = "http://localhost:11434";
    };
  };
in
{
  programs.zed-editor = {
    enable = true;
   # Zed-extensies (taal/LS/etc). "Continue" bestaat (nog) niet als Zed-extensie.
   # Dit is de officiële HM interface voor Zed extensions.
    extensions = [
      "nix"
      "toml"
      "rust"
      "org-mode"
    ];
    # Zed AI: Ollama als provider
    # Zed kan modellen auto-discoveren die jij met Ollama gepulld hebt.
    userSettings = {
      language_models = {
        ollama = {
          api_url = "http://localhost:11434";
          auto_discover = true;
          # Optioneel: zet een grotere context voor alle Ollama modellen
          # (Zed stuurt dit als `num_ctx` naar Ollama)
          context_window = 8192;
        };
      };
    };
  };
  # Continue config.json neerzetten (voor Continue in VS Code / JetBrains)
  # Pad: ~/.config/continue/config.json
  xdg.configFile."continue/config.json".text = continueConfigJson;
}