IncognitoPilot provides a local AI code interpreter for sensitive data analysis, leveraging models like GPT-4, Code Llama, or Llama 2. It empowers users to execute Python code, convert files, access the internet, and generate visualizations without uploading proprietary data to the cloud, offering a privacy-focused alternative to cloud-based solutions like ChatGPT Code Interpreter.

How It Works

IncognitoPilot integrates a Large Language Model (LLM) with a local Python interpreter. Users interact via a web UI, submitting natural language requests. The LLM generates Python code, which is then executed locally within a sandboxed environment. Approved code results are sent back to the LLM for context, enabling multi-turn interactions and self-correction. This local execution ensures data privacy, while the option to use open-source models like Code Llama allows for fully on-premises operation.

Quick Start & Requirements

• Installation: Docker is the primary method.

  docker run -i -t \
    -p 3030:80 \
    -e OPENAI_API_KEY="sk-your-api-key" \
    -e ALLOWED_HOSTS="localhost:3030" \
    -v /home/user/ipilot:/mnt/data \
    silvanmelchior/incognito-pilot:latest-slim
  

• Prerequisites: Docker, OpenAI API key (or configuration for local LLMs), a working directory for data access.
• Dependencies: The -slim image has minimal packages; remove -slim for a larger image with common data science libraries (e.g., for image/Excel processing).
• Links: Demo: https://github.com/silvanmelchior/IncognitoPilot/assets/6033305/05b0a874-6f76-4d22-afca-36c11f90b1ff

Highlighted Details

• Local execution of Python code for sensitive data processing.
• Supports both cloud APIs (GPT-4, GPT-3.5) and local open-source LLMs (Code Llama, Llama 2).
• Interactive UI with code execution approval and result display.
• Internet access for data retrieval and other tasks.
• Customizable dependencies via Dockerfile.

Maintenance & Community

• Active development by silvanmelchior.
• Contribution guidelines available for development and non-Docker installations.

Licensing & Compatibility

• License details are not explicitly stated in the README. Compatibility for commercial or closed-source linking is not specified.

Limitations & Caveats

The default latest-slim Docker image lacks common Python packages, requiring users to switch to the larger image or build custom images for tasks like image or spreadsheet manipulation. While data remains local, conversation history and approved code results are sent to cloud LLM APIs when used.