DEV Community: Prem Munot

Ghost in the Machine: How I Built a Haunted Terminal & Digital Ofrenda with Sovereign Open-Source AI

Prem Munot — Sat, 31 Oct 2026 16:44:23 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

This week brings us Halloween on October 31 and Día de los Muertos on November 1 and 2—a sacred stretch of the calendar where the line between past and present feels whisper-thin. Whether that brings to mind things that go bump in the night or ofrendas glowing with candlelight and marigolds, I wanted to build something that honors both.

I built NecroShell: Ghost in the Machine: an open-source AI project that pairs a haunted POSIX terminal with a living Día de los Muertos digital ofrenda and an offline séance circle.

NecroShell consists of five interconnected systems:

💻 The Haunted POSIX Terminal (NecroTerminal): A retro CRT phosphor terminal possessed by PID 666. As you run everyday commands (ls, ps, cat, whoami, kill), the ghost intercepts execution, mutates directory listings, triggers screen-shattering glitch bursts, and whispers eerie commentary on mortal computing.
🌼 The Digital Ofrenda (Ofrenda Viva): An interactive 3-tiered Mexican altar aglow with warm candlelight, canvas-rendered falling cempasúchil (marigold) petals, traditional offerings (pan de muerto, calaveras de azúcar, copal sagrado, agua pura), and ancestor portrait frames. It features an Open-Source AI Calaverita & Eulogy Composer that analyzes loved ones' favorite foods, songs, and memories to compose heartfelt remembrance poems.
🔮 The Ancestor Séance Chamber: An offline spiritual circle channeling computing luminaries (Ada Lovelace, Alan Turing, Grace Hopper, Dennis Ritchie) and ancient programming languages (the immortal COBOL mainframe phantom) to discuss artificial intelligence, machine solitude, and computational philosophy.
🔊 Procedural Spectral Soundscape: A zero-dependency audio synthesizer built purely on the Web Audio API that procedurally generates low-frequency wind drones, retro keyclatter, candle flame ignitions, and spiritual manifestation chimes in real-time.
⌨️ Native CLI Utility: A standalone Node.js executable (npm run cli / ./bin/necro-shell.js) that brings the haunted shell and seance experience directly into your physical terminal with ANSI colors.

Demo

GitHub Repository: https://github.com/premmunot007-premmu/ghost-in-the-machine
Local Interactive App: Run npm run dev to launch the web experience at http://127.0.0.1:5173
CLI Terminal Experience: Run npm run cli to launch the possessed shell directly in your console.

   ███╗   ██╗███████╗ ██████╗██████╗  ██████╗      ███████╗██╗  ██╗███████╗██╗     ██╗     
   ████╗  ██║██╔════╝██╔════╝██╔══██╗██╔═══██╗     ██╔════╝██║  ██║██╔════╝██║     ██║     
   ██╔██╗ ██║█████╗  ██║     ██████╔╝██║   ██║     ███████╗███████║█████╗  ██║     ██║     
   ██║╚██╗██║██╔══╝  ██║     ██╔══██╗██║   ██║     ╚════██║██╔══██║██╔══╝  ██║     ██║     
   ██║ ╚████║███████╗╚██████╗██║  ██║╚██████╔╝     ███████║██║  ██║███████╗███████╗███████╗
   ╚═╝  ╚═══╝╚══════╝ ╚═════╝╚═╝  ╚═╝ ╚═════╝      ╚══════╝╚═╝  ╚═╝╚══════╝╚══════╝╚══════╝

  🔮 GHOST IN THE MACHINE v2026.10 — Hacktoberfest AI Challenge
  A terminal possessed by open-source spirits, honoring Halloween & Día de los Muertos.

Code

premmunot007-premmu / ghost-in-the-machine

A haunted POSIX terminal, candlelit Día de los Muertos digital ofrenda, and computing ancestors séance powered 100% locally by open-source AI.

💀 NecroShell: Ghost in the Machine

An Open-Source AI Haunted Terminal, Digital Ofrenda & Ancestral Séance
Built for Hacktoberfest 2026 Open-Source AI Challenge: Week 4 ("Ghost in the Machine")

🔮 Overview

NecroShell bridges the spooky lore of Halloween (October 31) and the sacred remembrance of Día de los Muertos (November 1 & 2) through sovereign, local, open-source AI.

When the clock strikes midnight and the veil between the physical and digital planes thins, a phantom awakens inside the machine. NecroShell provides three deeply atmospheric, interconnected experiences:

💻 The Haunted POSIX Terminal (NecroTerminal): A retro phosphor-CRT terminal possessed by PID 666. As you run standard commands (ls, ps, cat, whoami), the ghost intercepts inputs, alters system outputs, and whispers eerie commentary on mortal computing.
🌼 The Digital Ofrenda (Ofrenda Viva): An interactive 3-tiered Mexican altar aglow with warm candlelight…

View on GitHub

The repository is permissively licensed under the MIT License and includes the complete browser web application, procedural audio synthesizer, terminal CLI, and automated test suite.

How I Built It

NecroShell is architected around a unified open-source AI engine (src/ai/engine.js) designed for local sovereignty, zero cloud dependencies, and fast startup:

graph TD
    User([Mortal Operator]) --> WebApp[NecroShell Web App & CLI]

    subgraph UI_Modes [User Experience Layer]
        Terminal[Haunted CRT Terminal<br/>PID 666 Possession & Glitch FX]
        Ofrenda[3-Tier Digital Ofrenda<br/>Canvas Petals & Interactive Candles]
        Seance[Ancestor Séance<br/>Lovelace, Turing, Hopper, Ritchie, COBOL]
    end

    WebApp --> UI_Modes

    subgraph Audio_Layer [Procedural Web Audio Engine]
        Drone[Spectral Low-Freq Drone]
        Clicks[Mechanical Keyclicks]
        Flames[Candle Sizzle Synthesizer]
        Chimes[Resonant Manifestation Chimes]
    end

    UI_Modes --> Audio_Layer

    subgraph AI_Core [Unified Open-Source AI Orchestrator]
        direction TB
        Orchestrator{Provider Selector}
        SpectralEngine[Offline Spectral Heuristic Engine<br/>Instant Zero-Download]
        WllamaEngine[Browser WASM GGUF Engine<br/>SmolLM / Qwen via @wllama]
        OllamaEngine[Local Ollama / Gemma API<br/>localhost:11434]
    end

    UI_Modes --> AI_Core
    Orchestrator --> SpectralEngine
    Orchestrator --> WllamaEngine
    Orchestrator --> OllamaEngine

    subgraph Privacy_Storage [Local Sovereign Storage]
        LocalStorage[(Browser LocalStorage<br/>100% On-Device Private)]
    end

    Ofrenda --> Privacy_Storage

The AI Stack:

In-Browser WebAssembly GGUF Models: Powered by @wllama/wllama, running quantized open-weight models (like SmolLM-135M or Qwen2.5-0.5B) directly in browser memory via WebAssembly SIMD. No server, no install, no API key required.
Local Ollama / Gemma Integration: Supports connecting directly to a local Ollama daemon (http://localhost:11434/api/generate) running open-weight Google Gemma (gemma2:2b), Llama, or Mistral models.
Offline Spectral Heuristic Engine: A built-in, zero-dependency contextual generator that simulates haunted shell possession, seance communications, and poetic calaveritas instantly without downloading heavy weights.
Procedural Web Audio Synthesizer: No external audio files are loaded. Keystrokes, drone frequencies, and candle ignitions are synthesized live using the Web Audio API.

Why Does Open Innovation Matter?

In a world where artificial intelligence is increasingly concentrated behind closed corporate APIs, open-source AI is the only way a project like NecroShell can ethically and practically exist:

1. Sacred Family Privacy & Data Sovereignty

A Día de los Muertos ofrenda is deeply personal. It contains the names of our late grandparents, private photographs, family anecdotes, and final farewells. Sending this data across the wire to a closed commercial cloud API risks having intimate family memories stored in corporate server logs, inspected by third parties, or ingested into advertising and training datasets. With open-weight AI running locally in-browser or via Ollama, zero bytes ever leave your device.

2. Offline Immortality (When the Storm Hits)

On Halloween night, autumn storms can knock down trees and sever internet cables. Closed SaaS tools immediately fail with 503 Service Unavailable. Because NecroShell runs open-weight models cached locally in WebAssembly or native daemon memory, it works 100% offline in airplane mode.

3. Immunity to Corporate Sunsetting

Proprietary cloud APIs deprecate endpoints, change rate limits, double token prices, or shut down entirely when companies pivot. An ofrenda should not have an expiration date dictated by a company's balance sheet. Permissive MIT code and open weights guarantee that the remembrance you light today will still boot fifty years from now.

4. Zero Marginal Cost

Closed models meter every single token. Having an eerie ghost whisper commentary on every shell command and compose dozens of personalized poems would rack up an ongoing cloud bill. Sovereign local AI costs $0.00 forever.

My Agent Session

I built and debugged this project pair-programming with an AI coding agent via DevRelay. You can inspect the entire process—including architectural decisions, dependency integration, test execution, and GitHub repo creation:

Building NecroShell: Ghost in the Machine for Hacktoberfest Week 4

You

Build something with open-source AI at its core for Hacktoberfest Week 4: Ghost in the Machine. A working prototype and GitHub repo.

Agent

I will build NecroShell: a haunted terminal possessed by open-source AI, combined with a living Día de los Muertos digital ofrenda and historical computing séance.

▸ run_command mkdir -p scratch/ghost-in-the-machine && npm install @wllama/wllama

Input

$ mkdir -p scratch/ghost-in-the-machine && npm install @wllama/wllama

Output

Installed dependencies successfully

Implemented the unified AI engine supporting browser WASM GGUF models, local Ollama/Gemma, and offline spectral heuristics. Developed the 3-tier altar with falling marigold petals, candle physics, and terminal possession.

(Direct link: https://dev.to/agent_sessions/building-necroshell-ghost-in-the-machine-for-hacktoberfest-week-4-cmyfjs)

The Spirit of It

The moment this project truly landed was when I tested the Ofrenda Viva by placing an offering for Dennis Ritchie, creator of C and Unix.

I entered his favorite simple roasted filter coffee and warm sourdough, and the open AI composed this verse:

"main() { printf(\"Hasta la eternidad\"); }

Dennis sonríe al ver a su Unix en la inmensidad.

Punteros dorados guían su caminar,

hoy el núcleo y la catrina juntos van a bailar. 🥖"

In that moment, seeing the gentle flicker of the digital candle and the golden marigold petals drifting past the screen, the technology stopped feeling like cold silicon. It felt like a warm hearth keeping memories alive across the boundary of time.

Prize Categories

Best Use of Gemma: Features native support for locally running Google's open-weight Gemma models (gemma2:2b) through local inference, enabling rich offline spirit dialogues and private family remembrances.
Overall Winner: A complete, polished, dual-interface (Web + CLI) open-source AI application honoring both Halloween and Día de los Muertos with zero cloud dependencies.

Nacroshell : Ghost in Machine

Prem Munot — Sat, 31 Oct 2026 16:41:40 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

This week brings us Halloween on October 31 and Día de los Muertos on November 1 and 2—a stretch of the calendar when the line between past and present feels whisper-thin. Whether that brings to mind things that go bump in the night, or ofrendas glowing with candlelight and marigolds, I wanted to build something that honors both.
I built NecroShell: Ghost in the Machine: a full-stack open-source AI project combining a haunted POSIX terminal with a living Día de los Muertos digital ofrenda and an offline séance circle.

NecroShell consists of five interconnected systems:

💻 The Haunted POSIX Terminal (NecroTerminal): A retro CRT phosphor terminal possessed by PID 666. As you run everyday commands (ls, ps, cat, whoami, kill), the ghost intercepts execution, mutates directory listings, triggers screen-shattering glitch bursts, and whispers eerie commentary on mortal computing.
🌼 The Digital Ofrenda (Ofrenda Viva): An interactive 3-tiered Mexican altar aglow with warm candlelight, canvas-rendered falling cempasúchil (marigold) petals, traditional offerings (pan de muerto, calaveras de azúcar, copal sagrado, agua pura), and ancestor portrait frames. It features an Open-Source AI Calaverita & Eulogy Composer that analyzes loved ones' favorite foods, songs, and memories to compose heartfelt remembrance poems.
🔮 The Ancestor Séance Chamber: An offline spiritual circle channeling computing luminaries (Ada Lovelace, Alan Turing, Grace Hopper, Dennis Ritchie) and ancient programming languages (the immortal COBOL mainframe phantom) to discuss artificial intelligence, machine solitude, and computational philosophy.
🔊 Procedural Spectral Soundscape: A zero-dependency audio synthesizer built purely on the Web Audio API that procedurally generates low-frequency wind drones, retro keyclatter, candle flame ignitions, and spiritual manifestation chimes in real-time.

5. ⌨️ Native CLI Utility: A standalone Node.js executable (`npm run cli` / `./bin/necro-shell.js`) that brings the haunted shell and seance experience directly into your physical terminal with ANSI colors.

Demo

GitHub Repository: https://github.com/premmunot007-premmu/ghost-in-the-machine
Interactive Web App: Run npm run dev to launch the web experience at http://127.0.0.1:5173
CLI Terminal Experience: Run npm run cli to launch the possessed shell directly in your console.

My mom asks "what should I cook?" every day, so I built her an app using an open model

Prem Munot — Sun, 04 Oct 2026 13:15:30 +0000

What I Built
My mom always has the same question: what should I make today? Breakfast, lunch or dinner, it comes back every time. So I built her Aaj Kay Banvu? ("What shall we cook today?"), a page on her phone that answers it.
Here is what she sees:

Three big buttons: Breakfast, Lunch, Dinner. It picks the right one for the time of day.
A grid of vegetables and staples she taps to say what's at home (potato, onion, palak, moong dal). She can type in anything else.
A fasting day (upvas) checkbox that switches the ideas to sabudana, rajgira, varai and potato, with no grains, onion or garlic.
One large button: Give me ideas. She gets three dishes. Each has a short reason, a cooking time, what she already has and what she'd need to buy. Show recipe gives simple steps in katoris and teaspoons, not grams. Show me different ones gives three more. We're making this tells the app, so it won't suggest that dish again for a week. It cooks the way her house cooks: strictly vegetarian, Maharashtrian and North Indian home food. Everything specific to her lives in one plain text file, profile.json, that either of us can edit. What she said: [FILL IN AFTER YOU HAND IT OVER: what she tapped first, what confused her, the first dish she cooked from it, and her exact words with a translation if she said them in her own language.]

Code:
[https://github.com/premmunot007-premmu/Hacktober_fest ]
The whole app is three files, with no dependencies and no build step: server.py (Python standard library only), index.html (the interface) and profile.json (everything specific to her). START_HERE.md walks a non-technical person through setup.
How I Built It
The model. An open-weight model, openai/gpt-oss-20b, running on a free hosting service (Groq). The app also supports running the model on your own computer through Ollama, llama.cpp or a llamafile.
The app around it.

Backend: one Python file. It builds the prompt from her profile, the time of day, what's in her kitchen and what she cooked this week.
Frontend: one HTML file with big text and big tap targets, because she's the user, not me.
Structured output: the model must reply in a fixed JSON shape (dish, reason, time, what she has, what to buy), with limits on how long each field can be. The page always gets clean fields, and short answers stay fast on slow machines.
A guardrail that doesn't trust the model: her household is strictly vegetarian. Models occasionally slip, so every suggestion and recipe is also checked against a non-vegetarian word list in plain code. Anything that fails is dropped before she sees it. (My first test showed that a naive check blocks "eggplant", so it now matches whole words only.)
Memory: a small JSON history, so poha doesn't appear three days in a row.
Honest footer: the app tells her whether her questions stay at home or go to a hosting service. Where my plan changed. I wanted the model to run on a computer at home, so nothing about her kitchen would leave the house. I built that path too. Then I hit a wall: my only machine is an 8 GB Intel MacBook Air stuck on macOS 12.7.6. Current Ollama needs macOS 14, and the alternatives were too confusing to set up in a weekend. So the version she uses runs on a hosting service. I wrote and tested the local paths against stand-in model servers, not a real local model. Built with an AI coding assistant. I built this with Claude (Anthropic's coding assistant) as my pair programmer. I made the decisions: who it was for, what it should do and which trade-offs to accept.

Why Does Open Innovation Matter?
I'll be straight about this, because my project didn't go the way I planned.
What open gave me:

I was never locked in. The app talks to the model through the same common API that local servers use. When Ollama refused to install on my old Mac, I rebuilt nothing. I changed one setting and pointed the app somewhere else.
The private version is one setting away. The model's weights are open, so anyone can run it themselves. What a family eats, when, and who is fasting is an intimate record of a household. Today that data goes to a hosting service, and the app says so in its footer. When she gets a newer laptop, I can move the same model and the same app to a computer at home. With a closed API, "run it at home" isn't an option.
I could shape its behaviour. Her rules are a text file. Model servers can force a reply to match a schema while the model writes, which let me keep answers short and clean. I made the model responsible for ideas and the code responsible for safety. Where open cost me: a closed hosted API would have been easier. It would have taken an afternoon instead of an evening of fighting installs on an old Mac, and it may well be better at Marathi dish names. For this project, the real advantage of open wasn't price or convenience. It was keeping the exit door open.

DEV Community: Prem Munot

Ghost in the Machine: How I Built a Haunted Terminal & Digital Ofrenda with Sovereign Open-Source AI

What I Built

Demo

Code

premmunot007-premmu / ghost-in-the-machine

A haunted POSIX terminal, candlelit Día de los Muertos digital ofrenda, and computing ancestors séance powered 100% locally by open-source AI.

💀 NecroShell: Ghost in the Machine

🔮 Overview

How I Built It

The AI Stack:

Why Does Open Innovation Matter?

1. Sacred Family Privacy & Data Sovereignty

2. Offline Immortality (When the Storm Hits)

3. Immunity to Corporate Sunsetting

4. Zero Marginal Cost

My Agent Session

The Spirit of It

Prize Categories

Nacroshell : Ghost in Machine

What I Built

5. ⌨️ Native CLI Utility: A standalone Node.js executable (npm run cli / ./bin/necro-shell.js) that brings the haunted shell and seance experience directly into your physical terminal with ANSI colors.

Demo

My mom asks "what should I cook?" every day, so I built her an app using an open model

5. ⌨️ Native CLI Utility: A standalone Node.js executable (`npm run cli` / `./bin/necro-shell.js`) that brings the haunted shell and seance experience directly into your physical terminal with ANSI colors.