gubbi

A minimal, chat-native LLM client for people who work in terminals and want explicit control over file attachments, session files, and provider switching without learning a plugin system.

Installation

pip install gubbi

Configuration

All configuration is stored in ~/.config/gubbi/settings.json. At least one provider needs added in order for chat to work. Either add it manually or program will ask for inputs on the first start.

{
    "providers": {
        "github": {
            "slug": "github",
            "url": "https://models.github.ai/inference",
            "default_model_id": "openai/gpt-4.1-mini"
        }
    }
}

In theory, it should work with any provider who follows openai api spec though I've only tested with github and few others so far. Let me know through issue tracker if it works for others.

Usage

usage: gubbi [-h] [-a] [-v]

Gubbi chat interface

optional arguments:
  -h, --help          show this help message and exit
  -a, --add-provider  Add OpenAI compatible provider
  -v, --version       show program's version number and exit

Commands

When a message starts with / (slash), it is interpreted as special command rather than a chat message sent to the LLM.

/help => List commands
/exit => Quit chat
/use <provider> => Switch provider
/model <model_name> => Switch model
/attach <path> => Attach a file
/clear => Clear context
/save => Save current chat
/load <filename> => Load an earlier chat
/models <filename> => List all models
/providers <filename> => List all providers	

Screenshot