Knowing Machine

The machine that knows the answers to all questions, using Interlaced Thinking (Kimi-k2) and a local knowledge base.

Features

• Interlaced Thinking: Uses the Moonshot Kimi-k2-thinking model to reason about problems.
• Local Knowledge Base: Integrates with a local LLM (e.g., via Ollama/LocalAI) to query private data.
• Rich Terminal UI: Displays the thinking process and final answer beautifully in the terminal.

Installation

pip install .

Usage

Prerequisites

1. Moonshot API Key: Set MOONSHOT_API_KEY environment variable.
2. Local LLM: Ensure a local LLM is running (e.g., Ollama on port 11434).

CLI

Run the machine from the command line:

python -m knowing_machine "What is the secret contained in my local notes?" --kb-model llama3

Python API

import asyncio
from knowing_machine import KimiClient, LocalLMKnowledgeBaseTool, InterlacedThinker

async def main():
    client = KimiClient() # Reads MOONSHOT_API_KEY env var
    kb = LocalLMKnowledgeBaseTool(model="llama3")
    
    machine = InterlacedThinker(client, [kb])
    await machine.think("Your complex query here")

if __name__ == "__main__":
    asyncio.run(main())

Configuration

• MOONSHOT_API_KEY: Required.
• Local Tool default base URL: http://localhost:11434/v1 (Compatible with Ollama).