TODO.md

LocalGPT - TODO

Dynamic System Prompt (Replacements)

• Config Model (src/Models/Config.php):

  • Add custom_fields property.
  • Add getCustomFields() method to retrieve the configuration.
  • Add setCustomFieldValue(string $key, string $value) to update/store field values at runtime.
  • In the constructor or a loading method, correctly parse the custom_fields array from gpt.json.

• Provider (src/Provider/BaseProvider.php):

  • Add a protected property protected array $customFieldValues = [];.
  • Add a public method setCustomFieldValues(array $values).
  • In setConfig(Config $config), call setCustomFieldValues() with any values from the config.
  • In buildSystemPrompt(), iterate through $customFieldValues and replace {{key}} placeholders in $this->systemPrompt with their corresponding values.

• Chat Command (src/Command/ChatCommand.php):

  • Argument Parsing:
    • Define a new --set option that accepts key-value pairs (e.g., --set pizza_region="New York"). This option should be repeatable for multiple fields.
    • In execute(), parse all --set arguments and store them.
  • Value Resolution Logic:
    • In execute(), after loading the GPT config, determine the final value for each custom field by checking in this order of priority:
      1. Value from --set command-line argument.
      2. Pre-defined value in gpt.json.
      3. Value from interactive prompt.
      4. default value from gpt.json.
  • Interactive Prompts:
    • For each custom field that still needs a value, prompt the user interactively.
    • Use Symfony Question Helper's Question for text type.
    • Use Symfony Question Helper's ChoiceQuestion for dropdown type.
  • Provider Integration:
    • Collect all final custom field values.
    • Pass the values to the provider instance using setCustomFieldValues().

LLPhant-Related Future Enhancements

• Add support for more providers via LLPhant
  • OpenAI
  • Anthropic
  • Groq
  • Ollama
  • DeepSeek
  • Grok
  • Grok 2
• Expose a public PHP API for using LocalGPT, with LLPhant as the backend.
• Create a "System Prompt Builder" agent using LLPhant to interactively define a GPT's persona.
• Implement reference file handling by using LLPhant's embedding and retrieval features.
• Implement chat history persistence using LLPhant's chat history features.