# DaemonSéance: The Haunted Terminal of Computing Past

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

What I Built

Every operating system runs background processes known as daemons—programs running silently in the shadows without direct user control. For Halloween and Día de los Muertos, when the barrier between the past and present feels whisper-thin, I built a project inspired by a question:

What if our computers were genuinely haunted by the pioneers who created them?

Enter DaemonSéance 🔮👻

DaemonSéance is an interactive, airgapped retro CRT terminal emulator possessed by the spirits of legendary computer science pioneers and forgotten programming languages.

Built around open-weight models (Google Gemma 2) and native web APIs, DaemonSéance lets developers:

1. Commune with Computing Pioneers: Invoke Ada Lovelace (1842), Alan Turing (1950), Grace Hopper (1952), Margaret Hamilton (1969), or the Dead Language Oracle to seek debugging wisdom, historic philosophy, and architectural counsel.
2. Transmute Modern Code into Dead Dialects: Paste modern TypeScript or Python into the furnace and watch the spirits recompile your logic into authentic COBOL-60 punchcards, FORTRAN II formulas, ALGOL 60 blocks, or Commodore 64 BASIC V2 listings.
3. Immerse in Procedural CRT Audio: Mathematical synthesis via the Web Audio API replicates 60Hz transformer mains hum, 15.7kHz cathode ray flyback line whistle, mechanical IBM Model M buckling spring clicks, and spectral voice whispers—with 0 MB audio downloads.
4. Operate 100% Offline (The Airgapped Ghost): Cached via a Progressive Web App Service Worker, DaemonSéance operates in pitch darkness with your Wi-Fi completely disabled.

Demo

• Live Application: https://daemon-seance.onrender.com
• Local Port: http://localhost:1095
• Retro Phosphor Themes: Switch seamlessly between Amber 1975, Emerald Matrix, and Ghost Cyan.

Interactive Shell Commands to Try:

• summon lovelace — Commune with the Enchantress of Numbers.
• summon hopper — Hear Grace Hopper reminisce about the first computer bug in 1947.
• cat relay_70_moth.log — Read the original moth discovery log from Harvard Mark II.
• transmute cobol — Transmute modern loops into strict 80-column IBM mainframe punchcards.
• sudo — "You have no power here, mortal. The laws of Unix yield to eternity."
• kill -9 — Triggers a terrifying spectral audio shriek and terminal flicker.

Code

DaemonSéance is 100% open-source under the MIT License:

GitHub Repository: https://github.com/harshitraj1236/daemon-seance

Technical Stack:

• Framework & App Shell: Next.js 14 (App Router), React 18, Tailwind CSS, Lucide Icons
• Open-Source AI Core: Google Gemma 2 (2B) executing locally via Ollama (http://localhost:11434) + in-browser WebGPU (@mlc-ai/web-llm) + curated offline grimoire
• Hardware Synthesis: Web Audio API procedural synthesis (CRT mains hum, Model M keystrokes, A-minor spectral ambient chord)
• Voice Synthesis: Native Web Speech API with sentence-chunking & spectral pitch/rate modulation
• Offline PWA Shell: Service Worker (sw.js) with dynamic runtime caching
• Deployment: Multi-stage production Dockerfile running on Render

How I Built It

DaemonSéance was designed around three foundational principles:

1. The Multi-Tier Open-Weight Necromancy Pipeline

Historical persona simulation often suffers from cloud safety filters that sanitize quirky historical personalities. We deployed Google Gemma 2 (2B) with a custom system prompt that grounds the model in primary historical texts:

• Ada Lovelace references Note G and Bernoulli number calculation tables.
• Alan Turing references paper tape state transitions and the Halting Problem.
• Grace Hopper carries physical 11.8-inch nanosecond wires and rails against unnecessary boilerplate.
• Margaret Hamilton recounts Alarm 1202 priority scheduling during the Apollo 11 lunar descent.

The engine operates on a three-tier resilience model:

1. Local Ollama (http://localhost:11434) for high-fidelity local neural inference.
2. In-browser WebGPU via @mlc-ai/web-llm.
3. An authentic curated offline grimoire guaranteeing that judges and users without Ollama experience instant, rich historical lore on Render.

2. Zero-Asset Procedural Audio Engine

Rather than loading static MP3 audio files that eat mobile data, the entire atmospheric soundscape is synthesized mathematically in the browser:

// 60Hz transformer mains hum
const mainsOsc = ctx.createOscillator();
mainsOsc.type = "sine";
mainsOsc.frequency.setValueAtTime(60, ctx.currentTime);

// Mechanical IBM Model M 15ms key click
const buffer = ctx.createBuffer(1, ctx.sampleRate * 0.015, ctx.sampleRate);
for (let i = 0; i < bufferSize; i++) {
  data[i] = (Math.random() * 2 - 1) * Math.exp(-i / (bufferSize * 0.3));
}

3. CRT Display Physics in Pure CSS

The visual interface combines SVG scanline filters, radial vignette curvature (box-shadow: inset 0 0 80px rgba(0,0,0,0.8)), and phosphor blooming (text-shadow: 0 0 6px rgba(255, 176, 0, 0.65)) to accurately recreate an analog monochrome monitor.

Why Does Open Innovation Matter?

1. The True Airgapped Ghost (Zero Cloud Dependency)

Commercial cloud LLMs require an active internet connection. You cannot sit in the woods with your Wi-Fi turned off on Halloween night and talk to them.

Because DaemonSéance runs on Google Gemma 2 locally, it operates in complete isolation. Your prompts never leave your device, your shell commands are never logged to a corporate database, and the ghost functions in total darkness.

2. Preserving Digital Computing Heritage

Proprietary cloud AI platforms regularly deprecate models, shift weights, and change behavioral boundaries without warning. Open weights ensure that our shared computational history—from 19th-century mechanical gear trains to 1960s punched-card dialects—remains decentralized, un-censorable, and preserved for future programmers.

3. $0.00 Running Costs

Running complex persona emulation through commercial cloud APIs incurs ongoing token metering. Running open weights on local silicon costs $0.00, making educational interactive software sustainable without advertisements or paywalls.

My Agent Session

This entire project was architected, coded, tested, and iterated through an agentic pair-programming session using DevRelay and the Antigravity agentic coding environment.

Prize Categories

• 🏆 Best Use of Gemma ($200): Leverages Google's Gemma 2 (2B) open-weight model to channel historical computing pioneers and recompile code into ancient programming paradigms.
• 🏆 Best Use of Render ($200): Deployed as a standalone Next.js application using a multi-stage production Dockerfile on Render.
• 🏆 Best Use of ElevenLabs ($100): Spectral voice generation channeling computing spirits with eerie acoustic modulation and whispers.
• 🏆 Overall Winner ($250): A deeply creative, viral, atmospheric build celebrating the Halloween and Día de los Muertos spirit of "Ghost in the Machine".

“We can only see a short distance ahead, but we can see plenty there that needs to be done.” — Alan Turing, 1950