GitGeist 👻: The Ghost in the Machines .week 4

GitGeist is an offline-first AI tool designed to analyze haunted Git histories and provide insights into suspicious code snippets. It's part of an Open-Source AI challenge for Hacktoberfest, aiming to help developers investigate old code without exposing private repository data. GitGeist combines two open-weight AI models, TabPFN and Google's Gemma 3, running on a local machine through Ollama.

The tool offers three main features. First is the TabPFN Haunting Forecaster, which analyses Git and system telemetry, including commit time, lines changed, cyclomatic complexity changes, days since a file was last modified, and test coverage signals to generate a Haunting Risk Score. Second is the Haunted Terminal Séance, which uses Gemma 3 locally to provide an understandable explanation of suspicious code.

This includes the commit hash, author, timestamp, and forecast from the risk engine. If Ollama is unavailable, a deterministic fallback can still format the available metadata. Finally, the Digital Code Ofrenda creates a digital remembrance card for retired code and its maintainers, giving developers a way to acknowledge past work.

GitGeist is designed for developers maintaining legacy codebases, engineers investigating regressions, open-source maintainers, and teams seeking local AI-assisted code analysis. The tool runs locally, including on CPU-only systems, without needing a paid AI API key. It offers two interfaces: a Haunted Terminal CLI and a web interface. The project's code is available on GitHub under the MIT license.