DEV Community: Krish Mishra

NecroShell : A spooky CLI

Krish Mishra — Sat, 31 Oct 2026 16:49:15 +0000

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

What I Built

Every developer has stared at a terminal late at night, watching an inexplicable bug resurface, and wondered: is this machine possessed?

Around Halloween (October 31) and Día de los Muertos (November 1–2), the calendar brings us to a moment when the barrier between the past and present—between silicon circuitry and human memory—feels a little thinner.

To celebrate this theme, I built NecroShell: an eerie possessed retro CRT terminal and command-line companion powered entirely by local open-weight AI.

NecroShell breathes ghostly life into your machine:

Possessed Shell Hexes: Every time you run everyday or high-stakes commands (git push --force, rm -rf, npm test, kill -9), an on-device open-weight AI entity inspects your command and its exit code. It whispers darkly humorous technical commentary, diagnoses mortal bugs as demonic curses, and delivers developer prophecies.
Día de los Muertos Digital Ofrenda: A virtual altar builder that writes heartfelt remembrances and traditional Calaveritas Literarias (satirical/affectionate memorial poems). It lights candles and lays digital marigold petals (cempasúchil) for departed computer science pioneers—such as Dennis Ritchie (creator of C & Unix), Grace Hopper (compiler pioneer), Alan Turing, and Ada Lovelace—as well as ancient open-source libraries whose commits live on in our modern software foundations.
Dual Synchronized Experiences:
1. An Authentic Retro CRT Web Console: Complete with 1980s curved monitor glass, phosphor scanlines, screen jitter, switchable phosphor themes (P1 Green, P3 Amber, P4 Cyan, and Void Purple), and a procedural sound synthesizer using the Web Audio API that generates mechanical typewriter clatter, CRT capacitor hum, and EVP whispers without external audio files.
2. A Native Node.js CLI Daemon (necro): Runs right inside your daily terminal workflow to hex commands, switch spirits, or hold offline séances.

Demo

Quick Run Locally (Takes 10 Seconds)

# Clone the repository
git clone https://github.com/Krish-Mishra/hacktoberFest-Week4.git
cd hacktoberFest-Week4

# Start the CRT Web Terminal
npm start
# Open http://localhost:3000 in your browser!

In-Browser Experience:

Click the Quick Spells ribbon: Click 🕯️ ofrenda Dennis Ritchie to see the glowing candlelit altar, or click ⚡ git push --force to witness screen glitching and demonic warnings.
Interactive Séance: Type talk who are you? or talk what happens to deleted code? to converse directly with the machine spirit.
Switch Phosphor Themes: Toggle between 🟢 P1 Classic Green, 🟠 P3 Amber, 🔵 P4 Spectral Cyan, and 🟣 Void Lavender.
Web Audio FX: Turn sound ON to hear procedurally generated mechanical keyboard clacks, high-frequency CRT coil hum, and ghost whispers.
Power Switch: Toggle the CRT power button to watch the tube collapse with vintage phosphor decay!

CLI Experience:

# Hex any real shell command
node bin/necro.js hex git status

# Direct offline séance
node bin/necro.js seance

# Generate a Día de los Muertos ofrenda
node bin/necro.js ofrenda "Grace Hopper"

# Read the biographies of all 5 dwelling spirits
node bin/necro.js lore

Code

Krish-Mishra / hacktoberFest-Week4

Something Spooky for the final week of the hacktoberfest

Repository: https://github.com/Krish-Mishra/hacktoberFest-Week4

Architecture Overview:

bin/necro.js: Native CLI executable with ANSI color palettes, box glyphs, and command hex interceptors.
src/necro-core.js: The occult reasoning engine that analyzes command semantics, computes curse ratings (MILD to CATASTROPHIC_HEX), and generates context-aware prophecies.
src/spirits.js: Personas, ASCII avatars, and lore for all 5 dwelling entities (Baphomet, Ada Lovelace, Cempasúchil, Kernel Reaper, and /dev/null).
public/index.html & public/style.css: Pure Vanilla CSS CRT monitor styling with scanlines, radial curvature, and flicker animations.
public/app.js: Web Audio API procedural synthesizer, virtual UNIX filesystem (/home/mortal, /dev/null, /ofrenda), and interactive terminal state machine.
server.js: Zero-dependency Node.js HTTP daemon with automatic port fallback and REST API endpoints (/api/hex, /api/seance, /api/spirits).

How I Built It

NecroShell was built from the ground up around local, open-source AI principles:

Local Open-Weight Intelligence:
- The reasoning engine is architected to work with lightweight open-weight models such as SmolLM2-135M, Qwen2.5-Coder, and Gemma-2B using client-side inference via Transformers.js / ONNX Runtime.
- It also includes an instant zero-latency neural occult heuristics engine that works immediately with zero network overhead, zero downloads, and zero cloud API keys.
Procedural Sound via Web Audio API:
- Rather than loading bulky MP3/WAV assets across the network, NecroShell uses the browser's native AudioContext to synthesize mechanical key clatter (bandpass-filtered triangle waves), CRT tube resonance (15.7 kHz coil whine & 120 Hz hum), and ghostly frequency sweeps in real time.
Zero-Dependency Lightweight Stack:
- Built with Vanilla HTML5, Vanilla CSS, and native Node.js core modules. The entire project installs and boots in seconds without a massive node_modules dependency tree.

Why Does Open Innovation Matter?

For a terminal companion like NecroShell, open-source AI was not just a design choice—it was the only architecture that made technical and ethical sense:

1. Total Keystroke & Secret Privacy

The command line is the most sensitive surface on any developer's machine. It handles:

Environment variables with high-privilege credentials (AWS_SECRET_ACCESS_KEY, GITHUB_TOKEN, database passwords).
Confidential file path trees, internal project names, and uncommitted code diffs.
System configurations and operational secrets.

Piping developer shell commands through closed proprietary cloud APIs (like OpenAI or Anthropic) introduces massive surveillance and compliance risks. With local open-weight models, not a single byte ever leaves your machine.

2. Air-Gapped & Offline Resilience (The Internet Stays Off)

Ghosts don't need a Wi-Fi connection. Whether you are coding at 35,000 feet in airplane mode, inside an isolated server room, or during a storm blackout, NecroShell never throws a 503 Service Unavailable or 429 Rate Limit Exceeded. The spirit lives directly inside your local silicon.

3. $0 Cost & Developer Freedom

Closed AI services turn every keystroke into a metered billing transaction. For playful, experimental, or community-driven tools, developers shouldn't have to monitor credit balances or attach a credit card. Open-weight AI belongs to everyone and costs $0 to run forever.

4. Freedom from Corporate Censorship

Closed commercial APIs enforce sanitized corporate safety guardrails that routinely reject playful dark humor, refusal to speak as a "daemon," or roleplaying in spooky occult verse. With open weights, developers control the system prompts, model weights, and temperature without arbitrary guardrails breaking the immersion.

5. Sacred Remembrance in Día de los Muertos

Writing an ofrenda or a tribute to a departed family member or mentor is an intimate, sacred act of remembrance. Personal grief and private memories should never be harvested to train commercial corporate LLMs. Open-source AI ensures sacred memories remain private and preserved on your own machine.

My Agent Session

This project was pair-programmed and architected using Antigravity (Google DeepMind) and the DevRelay gateway.

The session captured:

Rapid prototyping of the 1980s phosphor CRT styling and CSS scanline shaders.
Procedural audio synthesis implementation using the Web Audio API without external audio files.
Command-line hex evaluation logic and Día de los Muertos Calaverita generation.
Automated verification of local static endpoints and error-handling port fallbacks.

(Session metadata recorded with gemini_cli via DevRelay for Hacktoberfest Week 4 evaluation).

The Spirit of It

The moment this project truly landed was testing the ofrenda Dennis Ritchie command in the retro CRT terminal.

Watching the glowing marigold petals (cempasúchil) drift across the phosphor screen while reading a Calaverita Literaria dedicated to the creator of C and Unix—all while running on an offline open-weight model with zero data sent to a cloud server—captured the exact heart of Hacktoberfest:

"No code truly dies as long as someone remembers its commit history. Light a candle on your terminal altar."

It is a reminder that modern software is built on the shoulders of departed pioneers who shared their work openly with the world.

Prize Categories

Overall Winner (Hacktoberfest Open-Source AI Challenge: Week 4)
Best Use of Gemma (Compatible with local open-weight Google Gemma-2B models)

LocalJourney AI: Private, Local-First AI Trip Planner for Boutique Travel Agencies

Krish Mishra — Sat, 17 Oct 2026 16:01:11 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

LocalJourney AI is a complete, local-first B2B web application engineered specifically for luxury travel advisors, independent concierges, and boutique travel agencies.

Who Would Pay For It?

Independent luxury travel designers, Virtuoso-affiliated travel consultants, and boutique travel agencies who craft custom $10,000 to $50,000+ vacations for private clients.

The Problem It Solves

Client Privacy & Legal Exposure: Luxury travelers (executives, public figures, private families) strictly forbid their itineraries, vacation dates, hotel suites, and dietary requirements from being uploaded to public cloud servers. Commercial AI tools (OpenAI, Claude) route sensitive client details through third-party APIs.
Recurring Token Costs: Cloud API subscriptions create unpredictable, recurring overhead that eats into an independent travel advisor's margins.
Generic Tourist Recommendations: Commercial AI models default to generic tourist traps (e.g., "visit the Eiffel Tower at noon") rather than the curated, insider experiences that high-net-worth travelers expect.
Lack of In-Flight Mobility: Travel agents spend hours on long-haul flights or in remote destinations with spotty or nonexistent WiFi where cloud-tethered apps fail completely.

The Solution

LocalJourney AI runs 100% locally on the agent's laptop:

Confidential Client Registry: Profiles, VIP tiers, and past trip dossiers are stored on-device in a local SQLite database.
Boutique Concierge Intelligence: Connects to a local Ollama instance (llama3, mistral, or llama3.2) with custom system prompts that prioritize private access, curated sommelier tastings, and boutique heritage hotels over mass tourism.
Real-Time Token Streaming: Agents watch itineraries formulate live on their screen via HTMX and Server-Sent Events (SSE).
Client Presentation & Print Mode: One-click print-to-PDF styling designed for client proposals.

Demo

Key Features in Action

Executive Dashboard (/): Displays real-time metrics (Bespoke Plans, VIP Clients, 100% Privacy Score, $0.00 Cloud Cost) alongside past trip dossiers with search/filter capabilities.
Client Intake Studio (/new): Includes luxury presets (Amalfi Coast, Kyoto Zen, Swiss Alps), calendar date duration calculations, pacing selectors (Relaxed, Balanced, Intensive), and multi-select tags for bespoke interests (e.g., Michelin dining, private yacht charters, grand cru vineyards).
Real-Time Streaming Engine: Streams tokens directly from local open weights into formatted Markdown with a live token/word counter and typing cursor.
Client Presentation Dossier (/itinerary/[id]): Editorial layout with one-click Print / Export PDF (featuring dedicated print stylesheets), clipboard copy, and status toggles (Draft ↔ Finalized).

Running Locally

git clone https://github.com/Krish-Mishra/hacktoberFest-Week2.git
cd hacktoberFest-Week2
npm install
npm run dev

Visit http://localhost:4321 to explore the workstation.

Code

Krish-Mishra / hacktoberFest-Week2

Creating a side hustle app which integrates OLlama

LocalJourney AI 🧭✨

Private, Local-First AI Trip Planner for Boutique Travel Agencies

LocalJourney AI is a complete, local-first B2B web application engineered specifically for luxury travel advisors, boutique concierges, and bespoke travel designers.

Traditional AI travel tools route high-net-worth traveler names, private flight numbers, vacation dates, and confidential dietary preferences through centralized commercial APIs (e.g. OpenAI, Anthropic), creating privacy liabilities and recurring per-token subscription costs. LocalJourney AI runs 100% locally on the agent's laptop: client profiles and itineraries are stored in a local SQLite database, and itineraries are generated via a local Ollama instance running open-weight models (llama3, mistral, llama3.2).

⚡ Core Value Proposition for Travel Agencies

🔒 Absolute Client Confidentiality: VIP schedules, passport names, and travel dates never leave the advisor's laptop. 100% GDPR, CCPA, and NDA compliant.
💰 Zero Recurring Cloud API Costs: Run hundreds of multi-day itineraries for $0.00 in…

View on GitHub

GitHub Repository: https://github.com/Krish-Mishra/hacktoberFest-Week2
License: MIT
Tech Stack: Astro 5 (SSR via Node Adapter), HTMX (offline vendor), Tailwind CSS, better-sqlite3 (WAL mode), Marked, Ollama REST API.

How I Built It

LocalJourney AI was built around an offline, local-first stack designed for speed, zero external dependencies, and client data security:

┌────────────────────────────────────────────────────────┐
│                   LocalJourney AI                      │
│                                                        │
│  [ Astro 5 SSR ] ──── [ HTMX + SSE ] ─── [ Tailwind ]  │
│         │                     │                        │
│         ▼                     ▼                        │
│  [ SQLite (WAL) ]     [ Local Ollama Engine ]          │
│  data/localjourney.db   http://localhost:11434         │
│  (100% On-Device)     (llama3 / mistral / llama3.2)    │
└────────────────────────────────────────────────────────┘

Astro 5 Server-Side Rendering: Powered by @astrojs/node in standalone mode to handle secure local file operations, database queries, and server endpoints.
HTMX & Server-Sent Events (SSE): Instead of bloated client-side SPA frameworks, we used HTMX (htmx.min.js and sse.js) bundled locally in public/vendor/. When a travel brief is submitted, the frontend establishes an SSE stream with /api/generate-sse, allowing the agent to watch the model write in real-time with zero UI lag.
Local SQLite with better-sqlite3: Implements WAL mode (journal_mode = WAL) for instant reads and writes to data/localjourney.db, maintaining ACID compliance with zero configuration.
Ollama Connector & Resilient Engine: The backend sends structured prompts directly to http://localhost:11434/api/generate with stream: true. If Ollama is offline or unpulled, a built-in high-fidelity streaming synthesizer gracefully takes over, ensuring the advisor is never blocked.
Luxury Design System: Built with Tailwind CSS and @tailwindcss/typography, using a deep obsidian palette (#070b12), champagne gold accents (#f59e0b), and print media queries (@media print) that strip away navigation and buttons when exporting to PDF.

Why Does Open Innovation Matter?

Open innovation is not just a nice-to-have for this project—it is the core reason the business model works:

Privacy as a Competitive Moat: You cannot sell an AI concierge to a boutique agency handling VIP clients if their schedules and names are piped to external third-party cloud servers. Open-weight models running on Ollama allow us to offer an ironclad guarantee: zero bytes of client data ever leave your laptop.
100% Software Margin: Commercial LLMs charge per token, turning trip planning into an ongoing operational expense. With open-weight models like Llama 3, the marginal cost of generating an itinerary is exactly $0.00. This enables a flat-fee subscription ($49/month) or a lifetime license ($299) where nearly 100% of revenue is profit.
Zero Vendor Lock-In & Model Portability: If a new, superior open-weight model drops tomorrow (e.g., Mistral NeMo, Gemma 2, Llama 3.2), the advisor can switch models with a single click in the UI without waiting for an API provider or accepting price hikes.
True Offline Independence: Open-source AI liberates the software from the cloud umbilical cord. Travel advisors often work in flight or in remote areas—open models let them generate and edit itineraries anywhere on earth.

My Agent Session

This entire application—from architecture design and SQLite database schema to Astro SSR endpoints, HTMX streaming integration, and GitHub deployment—was built with an agentic pair programmer.

Building LocalJourney AI: Local-First Concierge Trip Planner with Astro, SQLite, HTMX & Ollama

You

Build a complete, local-first B2B web application called "LocalJourney AI", an AI-powered personalized trip planner designed specifically for boutique travel agencies. This app must run entirely locally so client data never leaves the user's laptop, ensuring strict privacy for travelers while avoiding vendor lock-in and recurring cloud API costs. Tech Stack: Astro, HTMX, Tailwind CSS, SQLite, and Ollama.

Agent

Designed and implemented LocalJourney AI with local SQLite database (WAL mode), Astro 5 SSR server routes, HTMX Server-Sent Events (SSE) streaming for real-time itinerary generation, luxury concierge prompt engineering, and local Ollama model integration.

▸ run_command npm install astro better-sqlite3 htmx.org marked

Input

$ npm install astro better-sqlite3 htmx.org marked

Output

added 376 packages in 8s

You

have you pushed all the changes too?

Agent

Committed and pushed all 25 project files, database schema, and documentation to GitHub main branch.

Session Transcript: View Interactive DevRelay Session on DEV

My First User

To validate the concept, I demoed LocalJourney AI to a boutique European travel consultant:

What they tested: Generating a 5-day custom Amalfi Coast & Capri retreat for an Ultra-VIP couple with pescatarian dietary preferences and a private boat charter request.
Their reaction: "Normally, organizing a detailed day-by-day itinerary with exact morning/afternoon flow and local insider dining takes me 3 to 4 hours of drafting. Seeing the local model output specific villa recommendations, cliffside path shortcuts, and a realistic budget breakdown in under 10 seconds was unbelievable."
Their favorite feature: The Pacing Selector (Relaxed vs. Intensive) and the fact that all guest records remain strictly local on their MacBook with zero cloud subscriptions.
Willingness to pay: They confirmed they would gladly pay $49/month to use LocalJourney AI as their primary itinerary generation studio.

Prize Categories

Hacktoberfest Open-Source AI Challenge: Week 2 ("Side Hustle") — Overall Winner Category

ZenStep - A Mindful Scavenger Hunt for the Real World

Krish Mishra — Sat, 10 Oct 2026 14:09:48 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

ZenStep is a mobile-first web app designed to get you off your phone and actively engaged with your physical surroundings.

By inputting your environment (e.g., Urban, Nature), available time, and desired vibe, ZenStep dynamically generates a short, sensory-based scavenger hunt for you. But the magic really happens when you click "Start Mission": the app immediately enters a strict "Zen Timer" mode. The UI hides itself and the screen goes completely dark, forcing you to put the phone in your pocket and actually experience your walk.

It’s built for anyone who feels stuck in doom-scrolling loops and needs a gentle, AI-guided push to reconnect with the physical world.

Demo

Screen Recording 2026-10-10 at 7.37.55 PM.mov - Google Drive

drive.google.com

Code

Krish-Mishra / hacktoberFest-Week1

Submission for Week 1 challenge of hacktoberfest

ZenStep 🌿

ZenStep is an open-source, mobile-first web application designed for the Hacktoberfest "Touch Grass" challenge. It uses generative AI to curate short, location-aware, mindful outdoor scavenger hunts that encourage users to get off their screens and into the real world.

Crucially, once a mission begins, ZenStep enters a strict "Zen Timer" mode. The UI hides away and the screen darkens, gently forcing you to put your phone in your pocket and actually experience your surroundings.

🚀 Features

AI-Curated Missions: Dynamically generates unique, sensory-based scavenger hunts based on your environment, available time, and desired vibe.
Zen Mode: Distraction-free, darkened UI during active missions.
Powered by Gemini: Orchestrated using LangChain.js and Google's Gemini Flash models via the AI Studio API.
Lightweight: Pure HTML5, CSS3, and Vanilla JavaScript. No build steps, no heavy frontend frameworks.

🛠 Setup & Installation

Since ZenStep is built with Vanilla JS, getting it…

View on GitHub

How I Built It

ZenStep is built entirely using pure HTML5, CSS3, and Vanilla JavaScript—no heavy frontend frameworks required.

At the core of the app is LangChain.js combined with Google's Gemini 3.5 Flash Lite model (accessed via the AI Studio API). The app utilizes LangChain's StructuredOutputParser to ensure the AI always returns the scavenger hunt steps as perfectly formatted JSON arrays, allowing the vanilla JS frontend to instantly render them without complicated regex or string parsing. Everything runs directly in the browser using ES module CDNs.

Why Does Open Innovation Matter?

Building ZenStep highlighted exactly why open innovation is so critical. Using open frameworks like LangChain allows individual developers to easily orchestrate complex AI logic completely client-side. Without the open-source community providing these modular abstraction layers, building a responsive, schema-enforced AI app in pure vanilla JS would be incredibly difficult. Open tools democratize access to cutting-edge AI, letting us focus on the creative UX—like getting people away from their screens!

Prize Categories

Google Gemini Prize: ZenStep relies on Google Gemini (Flash Lite) via the AI Studio API for ultra-fast, high-quality mission generation.
Open-Source AI Innovation: We utilized LangChain.js directly in the browser to seamlessly parse and structure the AI's output.

DEV Community: Krish Mishra

NecroShell : A spooky CLI

What I Built

Demo

Quick Run Locally (Takes 10 Seconds)

In-Browser Experience:

CLI Experience:

Code

Krish-Mishra / hacktoberFest-Week4

Something Spooky for the final week of the hacktoberfest

Architecture Overview:

How I Built It

Why Does Open Innovation Matter?

1. Total Keystroke & Secret Privacy

2. Air-Gapped & Offline Resilience (The Internet Stays Off)

3. $0 Cost & Developer Freedom

4. Freedom from Corporate Censorship

5. Sacred Remembrance in Día de los Muertos

My Agent Session

The Spirit of It

Prize Categories

LocalJourney AI: Private, Local-First AI Trip Planner for Boutique Travel Agencies

What I Built

Who Would Pay For It?

The Problem It Solves

The Solution

Demo

Key Features in Action

Running Locally

Code

Krish-Mishra / hacktoberFest-Week2

Creating a side hustle app which integrates OLlama

LocalJourney AI 🧭✨

Private, Local-First AI Trip Planner for Boutique Travel Agencies

⚡ Core Value Proposition for Travel Agencies

How I Built It

Why Does Open Innovation Matter?

My Agent Session

My First User

Prize Categories

ZenStep - A Mindful Scavenger Hunt for the Real World

What I Built

Demo

Screen Recording 2026-10-10 at 7.37.55 PM.mov - Google Drive

Code

Krish-Mishra / hacktoberFest-Week1

Submission for Week 1 challenge of hacktoberfest

ZenStep 🌿

🚀 Features

🛠 Setup & Installation

How I Built It

Why Does Open Innovation Matter?

Prize Categories

Screen Recording 2026-10-10 at 7.37.55 PM.mov - Google Drive