DevYard: Resurrecting the Ghosts in Your Code Machine with Local Open-Source AI

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
https://dev.to/challenges/hacktoberfest-week4-2026-10-26
What I Built

Every developer has a digital graveyard.

Tucked away in ~/projects, ~/code, or forgotten drive partitions lie dozens of abandoned repositories: hackathon
prototypes, late-night experiments, and side projects that quietly died when life got in the way. They sit dormant
on our hard drives—true "Ghosts in the Machine".

I built DevYard https://github.com/rajank18/devyard: a lightweight CLI tool and interactive digital cemetery that
performs forensic Git archaeology on your local code repositories. It unearths abandoned projects, extracts their
last living commits and unfinished TODOs, and unleashes local open-weight AI to channel their spirits:

• 🪦 Terminal Tombstones: Renders retro ASCII headstones with humorous AI-generated epitaphs and calculated causes
of death.
• 🕯️ Interactive Candlelit Graveyard: Generates an animated, responsive, dark-mode graveyard.html viewer complete
with flickering candle flames and project post-mortems.
• ⚡ Necromancy / Resurrection (devyard resurrect): Asks the AI spirit to evaluate the modern tech stack and
generate a step-by-step CPR revival plan, complete with an executable setup script to bring your dead project back
to life.
• 🔒 Zero Dependencies & Pure Offline: Zero npm runtime dependencies. Built entirely on native Node.js 18+ and
connects directly to local open-source models via Ollama.
Demo
You can run DevYard instantly via npx without installing anything globally:

# 1. Scan your local repos (runs fast, offline Git heuristics by default)
npx devyard scan

# 2. Unleash local Open-Source AI (e.g. Qwen 2.5 / Llama via Ollama)
npx devyard scan --ai

# 3. Generate the interactive candlelit graveyard webpage
npx devyard graveyard

# 4. Resurrect an abandoned project with an actionable revival plan
npx devyard resurrect my-abandoned-app

Terminal Output Preview:

Links:
• 📦 NPM Package: https://www.npmjs.com/package/devyard
• 💻 Source Code: https://github.com/rajank18/devyard
Code

rajank18 / devyard

A cemetery for abandoned side projects

🪦 devyard

A cemetery for abandoned side projects.

Catalog your unfinished repositories, commemorate their unfulfilled potential with terminal gravestones, and resurrect them when inspiration strikes again.

Runs 100% locally with zero external npm dependencies. devyard scan runs on pure local Git history with no AI required by default. If you want AI-generated epitaphs or revival advice, you can enable them with --ai or choose any provider (Ollama, OpenAI, Gemini, OpenRouter, Anthropic).

---

⚡ Quick Start

No installation required. Run directly with npx:

# 1. Scan your projects folder (zero AI, zero network calls, zero keys by default!)
npx devyard scan
# 2. Want creative AI epitaphs? Simply pass --ai or pick your preferred provider:
npx devyard scan --ai                   # Uses local Ollama
npx devyard scan --provider openai      # Uses OpenAI (OPENAI_API_KEY)
npx devyard scan --provider gemini      # Uses Google Gemini (GEMINI_API_KEY)
npx devyard scan --provider openrouter  

…

View on GitHub

Key architecture highlights:

• 0 External Runtime Dependencies: Everything is built using native Node 18+ APIs (node:fs, node:child_process,
node:http, node:crypto).
• AST / Regex Forensics: Extracts commit velocity, days since last breath, primary languages, and orphaned // TODO:
or // FIXME: comments across the codebase.
• Local Inference Engine: Implements a direct REST bridge to Ollama (http://127.0.0.1:11434), supporting local
models like qwen2.5:3b, llama3.2, and mistral.
How I Built It
DevYard is built specifically around local open-weight AI inference:

1. Local Open-Source Model (qwen2.5:3b / llama3.2 via Ollama): Instead of piping developers' private local file systems to closed external APIs, DevYard connects directly to Ollama. The model receives a compact context bundle containing the project's README header, top uncommitted TODOs, commit timeline, and latest commit messages.
2. Deterministic Fallback Engine: When running offline or without an active LLM daemon, DevYard uses a built-in heuristic rule-engine to calculate tombstone epitaphs from commit histories, ensuring the tool works anywhere—even with zero network access.

3. Pluggable Architecture:
   While optimized for local open-source models, DevYard's src/llm.js also supports multi-provider configurations
   (Ollama, OpenAI, Gemini, Anthropic, OpenRouter) with zero extra SDK bloat.
   Why Does Open Innovation Matter?
   Why does open innovation matter for a tool like DevYard?

4. Absolute Privacy for Unfinished Code:
   Side project directories contain unreleased prototypes, experimental ideas, and often accidental hardcoded tokens or
   private notes. Sending local source trees to closed third-party cloud APIs is a non-starter for privacy-conscious
   developers. With open-weight models running on Ollama, your code never leaves your machine.

5. Zero Cost & Infinite Shelf Life:
   Closed APIs introduce rate limits, paywalls, and breaking deprecations. Open models are permanent. You can run
   DevYard on an air-gapped laptop in a cabin with no internet, and it will run just as smoothly as in a datacenter.

6. Hackability:
   Open-source AI democratizes developer tooling. Anyone can fork the repo, tune the prompt templates in src/llm.js,
   switch to specialized quantized models, or add custom cemetery themes without needing an API billing account.
   The Spirit of It
   Every programmer carries a little bit of guilt over unfinished projects. We start with bursts of adrenaline,
   envisioning the next big product or tool, only to get distracted or burned out.

DevYard re-frames that guilt into nostalgia, humor, and closure. When the terminal prints an epitaph for a project
you haven't touched in 3 years, you don't feel bad—you laugh. And with devyard resurrect, that long-lost spark often
returns: you get a modern recap, a clean checklist, and the momentum to finally finish what you started.
Prize Categories
• Grand Prize / Open-Source AI Track
• Built with local open-weight inference (Ollama / Qwen / Llama), zero cloud dependencies, and 100% open-source code.