mull

A CLI thinking partner that critiques and refines its own responses.

Your first answer is rarely your best. mull finds the better one.

What it does

You ask a question. Instead of giving you a quick answer, mull thinks it through:

1. Initial response - Answers your prompt
2. Self-critique - Challenges its own answer using built-in "outsider principles"
3. Refined response - Delivers an improved answer informed by the critique

The result: more thoughtful, more honest, less shallow responses.

Install

git clone https://github.com/HCS412/local_llm_runner.git
cd local_llm_runner
pip install -e .

Setup

Start your local LLM server, then configure:

# LM Studio (default)
export MULL_API_BASE=http://localhost:1234/v1
export MULL_MODEL=your-model-name

# Ollama
export MULL_API_BASE=http://localhost:11434/v1
export MULL_MODEL=llama3

Or copy .env.example to .env and edit it.

Works with LM Studio, Ollama, llama.cpp server, vLLM, or any OpenAI-compatible API.

Usage

# Basic - shows refined answer only
mull "Should I quit my job to start a company?"

# Verbose - shows all thinking steps
mull -v "What's the best way to give difficult feedback?"

# Pipe input
echo "How do I handle a difficult conversation?" | mull

The principles

mull's critique is guided by built-in "outsider principles":

• Truth over comfort, especially when comfort belongs to the powerful
• Honor ambiguity - life is not binary, nor should thought be
• Cultural humility - never assume you see the whole story
• Speak with soul, not just safety
• Center the margins - not as charity, but as starting point
• Do not speak over - reflect alongside

These give mull its distinctive voice: direct, honest, willing to sit with complexity.

Requirements

• Python 3.9+
• A local LLM server running

License

MIT