This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
What I Built
What happens when a codebase is abandoned, a crash log is forgotten, or a terminal history becomes the only remaining record of a failed project?
I built GhostTrace AI, a local-first digital archaeology and forensic narrative platform that gives forgotten software projects a voice.
GhostTrace AI analyzes source files, TODOs, crash traces, and terminal histories to uncover technical clues and turn them into an interactive investigation. It separates its output into three layers:
Verified Evidence: Facts extracted by deterministic parsers, such as TODOs, panic messages, timestamps, and file locations.
Forensic Interpretation: AI-generated technical hypotheses that help explain what might have happened.
Spectral Narrative: Fictional, atmospheric storytelling from personas such as The Midnight Committer, The Panic Core, and The Deprecated Dependency.
The key idea is that evidence and imagination should coexist without being confused. Technical findings remain distinguishable from AI hypotheses and fictional storytelling.
I interpreted “Ghost in the Machine” as the digital traces left behind by software and the people who built it.
Demo
Youtube: https://youtu.be/3o1QL4CwEy8
GitHub repository: https://github.com/gunmasterg9/GhostTrace
Stable release: https://github.com/gunmasterg9/GhostTrace/releases/tag/v1.0.0
Code
Explore the source code, documentation, installation instructions, and release history:
GhostTrace AI on GitHub
GhostTrace AI v1.0.0 Stable Release
The project includes a React and Vite frontend, a local Ollama integration, evidence navigation, redaction safeguards, and Markdown dossier export.
How I Built It
GhostTrace AI is built around a local-first architecture using React, Vite, JavaScript, and Ollama for local language-model inference.
The application processes imported artifacts within explicit limits, extracts evidence, and presents investigation results in a structured interface. It can stream responses from a locally running Ollama model and fall back to an offline simulation engine when inference is unavailable or times out.
Models tested locally include:
qwen3:1.7b — continuous streaming verification.
gemma4:e4b — inactivity-timeout and offline-fallback verification.
The project also includes safeguards for sensitive text, interactive source citations, and inert parsing of terminal histories. Imported terminal commands are analyzed as text rather than executed.
The Windows verification report recorded 36 passing automated tests, a clean lint run, a successful production build, and zero vulnerabilities reported by npm audit at the time of testing.
Why Does Open Innovation Matter?
Open innovation makes it possible to experiment with AI without making a hosted, closed API the center of the architecture.
With local open-weight models through Ollama, GhostTrace AI can investigate software artifacts on a developer's own machine. This supports a privacy-focused workflow and gives developers more control over model selection, inference, and experimentation.
It also makes the project easier to explore, adapt, and improve with community contributions. The aim is not to claim that local inference eliminates every privacy risk, but to reduce dependence on remote AI services and make the system's boundaries more transparent.
Open innovation also encourages developers to question AI output, inspect the implementation, reproduce behavior, and improve the tools they rely on.
My Agent Session
An agent-session recording is not included in this submission yet.
If I publish a session through DevRelay, I will add the recording or embed it here so readers and judges can see how the project was built and tested.
The Spirit of It
Every abandoned repository contains a small mystery.
A TODO left unfinished. A panic message from the last run. A terminal command typed at 2 a.m. by someone trying to make the project work.
GhostTrace AI turns those traces into an investigation. The parser identifies what the files actually contain; the model proposes possible explanations; the narrative gives the investigation atmosphere.
The ghost is fictional. The evidence should be inspectable.
That distinction is the heart of the project: AI can help us tell a story about forgotten software without pretending that the story itself is proof.
Prize Categories
Primary challenge: Hacktoberfest Open-Source AI Challenge Week 4 — Ghost in the Machine.
Additional partner prize categories have not been confirmed. I will add only the categories for which the project meets the official eligibility criteria.
Top comments (0)