DEV Community: Suraj Dhawal

Ghost in the Machine: Possessed Terminal & Día de los Muertos Living Ofrenda with Open-Source AI

Suraj Dhawal — Sat, 31 Oct 2026 06:42:29 +0000

👻 What I Built

Between October 31 (Halloween) and November 1–2 (Día de los Muertos), the veil between past and present thins. I wanted to build an experience that honors both halves of this mystical season: the spooky cyber-folklore of machines possessed by historical memories, and the sacred, golden glow of ofrendas carrying the spirits of our ancestors and computing pioneers.

I built Ghost in the Machine — an offline-first cyber-folklore installation powered 100% by open-source AI, in-browser neural tokenization, procedural Web Audio synthesis, and local Ollama inference.

Key Features & Subsystems

1. NecroShell v3.13 (The Possessed Terminal & Séance)

A retro CRT green-phosphor shell possessed by computing history's greatest departed minds (Ada Lovelace, Alan Turing, Grace Hopper, Dennis Ritchie, and Hedy Lamarr).

Procedural CRT Keystrokes & Glitches: Authentic mechanical key clicks with pitch jitter and chromatic aberration scanlines.
Dynamic 100% Master Sound Level: Synchronized real-time volume percentage (100%, 70%, MUTED), interactive volume slider, and instant mute toggle.
Séance Invocations: Type seance ada, seance turing, whisper <query>, cat hopper, or possessed to channel the pioneers.
Organized Quick Spells Dock: 1-click grimoire pills to trigger séances, possess glitches, or jump straight to other sanctuaries.

2. Altar de Recuerdos (The Living Ofrenda)

A sacred Día de los Muertos sanctuary illuminated by flickering candlelight, traditional papel picado banners, and floating marigold (cempasúchil) petals.

Canvas Particle Physics Engine: Real-time dual-bezier falling marigold petals with rotational gravity and turbulent wind oscillation.
Solfeggio 528Hz Chimes: Light virtual candles with personal prayers accompanied by the resonant 528Hz "miracle frequency" harmonic chime.
Ancestral & Pioneer Memorials: Honor computing giants or add personal tributes for departed loved ones with persistent browser memory.

3. Calaverita Literaria & Digital Memorial Card Exporter

On-Device Folk Verses: Generates authentic Mexican Calaveritas Literarias (satirical, loving poetic rhymes where La Catrina visits the departed) and heartfelt digital eulogies celebrating favorite foods, songs, and memories.
HTML5 Canvas 1000×650 Card Exporter: Generates and downloads high-resolution PNG memorial cards complete with ornate Mexican double borders, marigold emblems, and gold flourishes.

4. The Phantom Investigator (IoT Radar & Telemetry Oscilloscope)

A 60fps radial radar canvas tracking simulated midnight smart home IoT sensor streams.

Infrasound Oscilloscope: Plots real-time waveforms for the 18.98Hz infrasound wave (the "ghost frequency") and 7.83Hz Schumann planetary resonance.
Multi-Format Export: One-click export of paranormal forensic logs into formatted .txt, structured .json, or tabular .csv.
Zero Cloud Leaks: Every anomaly is parsed locally without shipping sensor telemetry to commercial data brokers.

5. Royal Farman Scroll Apprentice Guide

An ancient Indian Maharaja Farman scroll aesthetic with carved gold dowels (✦), a 3D wax seal, and copyable decree cartouches.

Provides step-by-step guidance on what to type in NecroShell with 1-click copy buttons.
Explains the cultural significance of the Living Ofrenda and how to calibrate the Telemetry Radar.

6. Interactive Halloween Ghost Companions

Fluid Custom Ghost Cursor: Features a 0-latency target reticle paired with a smoothed trailing ghost companion driven by damped spring physics, blushing cheeks, eye look tracking, and anti-burst teleport clamping.
Animated Ghost Mascot ("Fantasma"): Hovers in the corner with gentle sine-wave floating and contextual folklore advice.

🌐 Demo & Code

Live Production URL: https://ghost-in-the-machine-kappa.vercel.app
GitHub Repository: https://github.com/Dev-Suraj-Dhawal/ghost-in-the-machine

Dev-Suraj-Dhawal / ghost-in-the-machine

👻 Ghost in the Machine: Offline Spectral Terminal, Possessed Shell & Día de los Muertos Living Ofrenda powered by Open-Source AI

👻 Ghost in the Machine

Offline Spectral Terminal, Possessed Shell & Día de los Muertos Living Ofrenda

Powered 100% by Open-Source AI, Web Audio Procedural Synthesis & Local Inference

"When the boundary between living carbon and eternal silicon grows razor thin."

🌐 Experience Live Séance on Vercel • 🏛️ Architecture Spec • 🛡️ Security Model • 🤝 Contributing Guide

📸 Visual Showcase & Feature Tour

Experience the core interfaces of Ghost in the Machine:

📟 1. NecroShell Possessed Terminal	🕯️ 2. Living Ofrenda Sanctuary Altar

Retro green-phosphor CRT shell with procedural audio synthesis, 100% synchronized dynamic master volume, and quick séance spells.	Candlelit Día de los Muertos sanctuary with floating dual-bezier marigold petal physics, custom eulogies, and Solfeggio 528Hz chimes.

📡 3. Phantom Investigator Telemetry Radar	🛡️ 4. Why Open Innovation Manifesto

60fps radial radar sweep tracking smart home anomalies, 18.98Hz infrasound, and 7.83Hz Schumann planetary resonance.	Architectural breakdown

…

View on GitHub

🛠️ How I Built It

1. Open-Source AI Inference Architecture

The core principle was zero closed APIs, zero external surveillance, and 100% offline autonomy.

In-Browser Local Neural Engine: A deterministic, client-side inference core running directly in the browser. It delivers instant, zero-latency séances and Calaverita poetry generation without requiring any API keys or network connection.
Local Ollama Daemon Bridge: Built-in support for users running local open-weight models via Ollama at http://localhost:11434. Seamlessly channels llama3.2, mistral, smollm2, gemma2, or qwen2.5 on the user's local machine.

2. Procedural Web Audio Synthesizer (0 External MP3 Files)

Instead of shipping heavy audio files, I programmed a procedural Web Audio API synthesizer in src/lib/soundEngine.ts:

Realistic mechanical CRT key clicks with randomized pitch jitter.
Eerie 55Hz sub-bass oscillator and filtered noise bursts for spectral possession glitches.
Solfeggio 528Hz ("miracle frequency") harmonic bell chimes when candles are lit.
Observable subscribeVolume architecture ensuring the master sound percentage is 100% synchronized across the entire application.

3. Marigold Canvas Physics Engine

To capture the warmth of Día de los Muertos, LivingOfrenda.tsx renders dynamic falling marigold petals (cempasúchil) using dual-bezier curve paths, rotational physics, and turbulent wind oscillation.

4. Technical Stack

Frontend: React 19, Vite 6, TypeScript 5.7
Styling: Tailwind CSS v4, custom CRT scanline shaders, phosphor glow filters
Deployment: Vercel Global Edge Network with strict Cross-Origin Isolation headers (COOP/COEP)
CI/CD: GitHub Actions automated typecheck and production build pipeline

🕯️ Why Does Open Innovation Matter?

In a world increasingly dominated by opaque, closed cloud APIs, building Ghost in the Machine with open innovation proved why open-source AI is not just an alternative — it is the only ethical choice for sensitive human experiences:

Sacred Grief & Data Sanctity: When a user writes an intimate remembrance of their departed grandmother, their words describe deep human love and vulnerability. Closed corporate models ingest user inputs into remote databases, tokenizing and retraining advertising algorithms on private tears. With our open architecture, zero bytes ever leave the user's laptop.
Resilience When the Lights Go Out: Halloween storms frequently cut power lines and internet connections. A séance that fails because of a 503 Service Unavailable error from a Silicon Valley server destroys the magic. With local open-weight models, the terminal speaks by candlelight in full offline isolation.
Eternal $0 Cost: Ancestral remembrance should never depend on a credit card subscription. A project running on open models can be preserved on a USB key or disk image and still boot 50 years from now.
Cultural Sovereignty: Closed APIs often sanitize cultural folk practices or flag gothic Halloween narratives as unsafe. Open weights give developers complete freedom to fine-tune models on authentic Mexican Calaveritas, regional idioms, and historical computing archives.

🤖 My Agent Session

This project was built collaboratively with Google DeepMind Antigravity, leveraging advanced agentic pair programming:

Autonomous iterative git feature branches and pull request workflows.
Full verification loops (tsc, vite build, and GitHub Actions CI pipelines).
Procedural audio synthesis and fluid canvas physics debugging.
Strict security hardening with Content-Security-Policy and ErrorBoundaries.

🌺 The Spirit of It

During testing, I typed seance ada into the possessed terminal. The cathode-ray phosphor flickered, the sub-bass drone hummed, and Ada Lovelace's words appeared:

"The Analytical Engine weaves algebraical patterns just as the Jacquard loom weaves flowers... User, what algorithm will you weave before the candlelight fades?"

Then I switched to the Living Ofrenda, added a tribute for my departed grandmother, and watched the AI compose a Calaverita honoring her warm kitchen and Sunday laughter while golden marigold petals drifted across the screen.

It gave me chills, and then it gave me warmth. That is the spirit of Día de los Muertos: the technology in our hands is simply the newest medium through which we keep our loved ones and heroes alive.

🏆 Prize Categories

Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine
Open Innovation & Local Inference Showcase

SlothFlow AI: The Slow & Steady Open-Source AI Suite That Rewards Laziness

Suraj Dhawal — Sun, 25 Oct 2026 15:14:34 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

What I Built

SlothFlow AI is a client-first, zero-subscription-cost suite of anti-effort utilities engineered for anyone who wants to channel their inner sloth on International Sloth Day.

Instead of demanding complex configurations or monthly SaaS subscriptions, SlothFlow AI tackles the exact micro-tasks people dread the most, letting you conserve mental energy:

Chaos to 2-Line Clarity (Chat & Inbox Tidier): Ingests 50+ unread messages from family or work group chats and extracts a 2-sentence "Sloth Brief", flags whether you actually have to do anything, and provides 3 instant 1-click polite replies.
Lazy Voice Ramble to Clean Roadmap: Lie back, record your half-asleep mumblings, and let local open models structure them into clean checklists, grocery items, or bug notes.
Fridge-to-Feast (No-Brain Cooking): Never stand in front of the open fridge staring into the void again. Enter leftover scraps and receive a 10-minute, 3-step meal that only soils 1 pan and 1 fork.
The Procrastination Breaker (Dreaded Reply): Generates guilt-free, tactful replies for emails and Slack DMs you've been delaying answering for two weeks.
Wind-Down & Snooze Sanctuary: An offline 4-7-8 breathing circle synchronized to Sloan the Sloth's resting heart rate, paired with slow-cadence bedtime micro-stories and an offline procedural audio synthesizer (rain, thunder, forest night breeze, brown noise) built with the Web Audio API.

Who gets to be lazy? Everyone. Students, tired developers, parents, and anyone experiencing digital burnout.

Demo

Live Application: https://slothflow-six.vercel.app
GitHub Repository: https://github.com/Dev-Suraj-Dhawal/slothflow

🖥️ Main Dashboard & Sloan Field Log

🎛️ Live Workspace & Procedural Soundscape Synth

Code

Dev-Suraj-Dhawal / slothflow

🦥 Slow & Steady open-source AI suite that rewards laziness. Built for Hacktoberfest 2026.

🦥 SlothFlow AI — Slow & Steady Open-Source AI That Rewards Laziness

🚀 Live Demo: https://slothflow-six.vercel.app

"Channel your inner sloth. Build something with open-source AI that does the work so you don't have to." Built in honor of International Sloth Day and Sloan the Sloth (DEV's mascot).

🌟 Overview

SlothFlow AI is a client-first, zero-server-cost productivity and relaxation suite powered entirely by open-weight models and local inference. Instead of paying monthly API fees or piping private chats and audio rambles to remote servers, SlothFlow runs open models directly in your browser (via WebGPU & Transformers.js) or connects to your local desktop Ollama daemon.

It exists for one simple purpose: to eliminate dreaded micro-tasks so you can conserve energy like a sloth.

📸 Interface & Architecture Tour

1. Main Dashboard & Energy Telemetry

Clean cyber-terminal UI featuring live energy telemetry (effort saved, calories stored, 0 KB egress), Sloan mascot field…

View on GitHub

How I Built It

🔌 Multi-Engine Open Inference Architecture

SlothFlow AI is built entirely around open-weight models and local inference architectures:

Engine A — In-Browser WebGPU Open Weights: Using @huggingface/transformers (Transformers.js v3) and ONNX Runtime Web, models like onnx-community/Qwen2.5-0.5B-Instruct and Xenova/distilbart-cnn-6-12 execute client-side directly on the user's GPU/WASM threads. Zero server calls; models are cached in the browser Cache API.
Engine B — Local Desktop Ollama Bridge: For developers with local GPUs, SlothFlow connects directly to localhost:11434 to stream completions from llama3.2, mistral, smollm2, or deepseek-r1.
Engine C — Instant Distilled Heuristics: Pure offline rules distilled from open models providing zero-latency fallback for low-power mobile devices.
Procedural Audio Engine: 100% offline procedural audio synthesizer utilizing the Web Audio API (white/brown noise buffers, BiquadFilters, and LFO modulators). No streaming MP3s, zero network requests.

Why Does Open Innovation Matter?

The challenge asked: “Why does open innovation matter for what you built? What did it make possible that a closed API wouldn't?”

Zero-Cost Sustainability ($0 Forever): A closed API model with pay-per-token metering penalizes users for mundane, everyday laziness tasks. Open weights allow users to automate low-stakes chores without subscription anxiety or credit card friction.
Sacred Bedroom & Personal Privacy: SlothFlow ingests raw, unedited personal data — awkward draft emails, chaotic family squabbles, half-asleep mumblings, and messy pantry scraps. Sending that telemetry to proprietary cloud servers presents unnecessary privacy risks. Running open-weight models locally ensures that not a single byte leaves the machine.
True Offline Independence: True relaxation often occurs when you intentionally unplug. SlothFlow functions on an airplane, in a remote cabin, or in a bedroom with Wi-Fi switched off. Closed APIs turn into a loading spinner the second connectivity drops.
Modularity Without Deprecation: Closed APIs change guardrails, drift in formatting, or get sunset on vendor schedules. Open weights are immutable, inspectable files that the user owns forever.

The Lazy Payoff

During testing this week, SlothFlow eliminated an estimated:

65 minutes of unproductive screen scrolling and decision fatigue.
135 kcal of mental energy that would have been wasted staring blankly into the fridge or dreading awkward emails.
4 messy dishes saved by cooking the 1-pan Sloth Skillet.

Prize Categories

Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady (Primary Track)

TerraForge Studio — Open-Source AI Terrarium CAD & Side-Hustle Engine

Suraj Dhawal — Sun, 18 Oct 2026 12:43:26 +0000

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

Empowering botanical micro-businesses, terrarium artisans, and plant side-hustlers with local open-weight AI, interactive vector CAD, and deterministic microclimate simulation.

What I Built

TerraForge Studio is a local-first, zero-cloud-fee computer-aided design (CAD) workspace, biological microclimate simulator, and commerce co-pilot built specifically for boutique terrarium artisans, micro-botanical shop owners, and plant craft side-hustlers.

Boutique living terrariums and mossariums are booming artisan craft products. Makers sell handcrafted Wardian cases, apothecary jars, and bio-active domes on Etsy, Shopify, and at weekend local farmers' markets for $65 to $160+ each. However, running a botanical micro-business involves three painful roadblocks:

Biological Trial-and-Error (The $40 Mistake): Improper substrate layering (e.g., missing an activated charcoal filtration barrier or having insufficient LECA drainage) leads to root rot, sour anaerobic odors, and dead plants within 14 days—triggering customer refund requests and lost reputation.

Wholesale Margin & Supplier Exposure: High-margin builders source rare bryophytes, tissue-culture cuttings (Marcgravia), and vintage glassware from niche wholesale distributors. Feeding raw Bills of Materials (BOM) or supplier invoices into closed third-party cloud AI tools exposes proprietary vendor pricing and business margins.

SaaS Subscription Fatigue: When an artisan craft margin is $40–$55 per terrarium, paying $29–$49/month for cloud AI tools and subscription software eats up a significant portion of their profits. Worse, weekend craft fairs, greenhouse basements, and pop-up markets rarely have dependable Wi-Fi.

The Solution

TerraForge Studio replaces clunky spreadsheets, handwritten care tags, and costly cloud APIs with a single unified, local-first studio:

Interactive SVG Vector CAD Canvas: Artisans pick glass vessels, adjust 5-layer substrate depths to the millimeter, and drag-and-drop living specimens (Fittonia, Cushion Moss, Maidenhair Fern, Dragon Stone, Springtail clean-up crews) with real-time scale, rotation, and layering controls.

Deterministic Bio-Compatibility Engine: Dynamically calculates relative humidity (RH%), grow-light lux, and ecosystem stability, instantly warning makers before they pair rotting desert succulents with high-humidity cloud forest flora.

Commercial Unit Economics & Margin Calculator: Automatically tabulates the raw Bill of Materials (glass + substrate millimeters + specimens), factors in artisan assembly labor, and outputs retail pricing and net profit margins.

Local Open-Weight AI Co-Pilot: With zero cloud API fees, it synthesizes high-converting Etsy SEO listings (with 13 algorithm tags), printable Botanical Care Passports for buyers, and pre-shipping pathology audits.

Demo

Live Deployed App (Vercel): https://terraforge-studio.vercel.app

GitHub Repository: https://github.com/Dev-Suraj-Dhawal/terraforge-studio

Local Quick Run: Clone the repository and execute python -m http.server 8080 (or npx serve .), then open http://localhost:8080.

Interactive Workflow Walkthrough

Choose Vessel & Stratigraphy: Select between a Geometric Wardian Case, Victorian Apothecary Jar, Erlenmeyer Bio-Flask, Botanical Bowl, or Gallery Cylinder. Adjust LECA drainage, activated charcoal, sphagnum barrier, ABG soil mix, and silica sand with instant millimeter precision.

Plant & Design in Vector CAD: Drag living flora and hardscape specimens into the glass container. Scale, mirror, and arrange the micro-landscape visually with zero latency.

Check Live Bio-Telemetry: Watch the real-time Viability Index (0–100%), internal humidity (RH%), and automated pathology alerts.

Tune Unit Economics: Adjust the assembly labor slider ($5–$45) and retail markup multiplier (1.5x–3.5x) to view real-time raw COGS, retail price, and net profit.

Generate Commerce Deliverables via Local AI: Click Generate to produce an Etsy SEO listing with 13 tags, a printable customer Care Passport card, or a pre-shipping audit—running locally at $0.00 marginal cost.

Export Blueprint: Export an architectural vector blueprint SVG with one click for workshop records and physical assembly.

Code

The complete source code is modular, zero-dependency, and open-source under the MIT license:

GitHub Repository: https://github.com/Dev-Suraj-Dhawal/terraforge-studio

Dev-Suraj-Dhawal / terraforge-studio

🌱 TerraForge Studio

Interactive Terrarium CAD, Real-Time Microclimate Bio-Simulator & Local Open-Weight AI Side-Hustle Operating System

📌 Executive Summary

TerraForge Studio is an all-in-one computer-aided design (CAD) workspace, biological compatibility engine, and automated e-commerce co-pilot created specifically for artisan terrarium makers, micro-botanical studio owners, and plant side-hustlers.

Boutique bio-active terrariums are high-margin, artisan products frequently sold at weekend farmers' markets, pop-up events, and craft marketplaces like Etsy and Shopify. However, builders face critical pain points:

Biological Trial-and-Error: Improper substrate layering (missing false bottoms, insufficient activated charcoal) causes root rot and ecosystem collapse.
Proprietary Margin Exposure: Handcrafters source rare botanical clippings and wholesale glassware; sending trade secrets and wholesale Bills of Materials (BOM) to cloud SaaS tools leaks private pricing.
Cloud Subscription Fatigue: A $29–$49/month recurring cloud AI subscription severely cuts into small artisan craft margins, and requires Wi-Fi in basements or outdoor weekend markets.
General LLM Hallucinations: Cloud LLMs often…

View on GitHub

Architecture Overview

index.html: Clean semantic single-page layout styled with Tailwind CSS and glassmorphic botanical aesthetics.

assets/js/catalog.js: Data models for vessel geometries, substrate physical depths, wholesale costs, and botanical biome specifications.

assets/js/state.js: Reactive application state store, canvas drag coordinates, and pre-configured presets (Zen Mossarium, Fittonia Rainforest, Bio-Conflict Test).

assets/js/vector-engine.js: Pure SVG 2.0 rendering pipeline with dynamic clip-paths, organic wave substrate calculations, specimen transforms, and vector export.

assets/js/bio-simulator.js: Deterministic biological rule validator calculating internal microclimates and cohabitation viability.

assets/js/economics.js: Dynamic Bill of Materials (BOM) cost engine and margin calculator.

assets/js/ai-copilot.js: Local Ollama bridge (http://localhost:11434/api/generate) and zero-fee offline structured JSON generation harness.

assets/css/style.css: Botanical color palette, glassmorphism filters, tactile slider controls, and ambient mist animations.

How I Built It

TerraForge Studio is engineered around open-source AI and offline-first computing:

Local Open-Weight Inference Engine

The core intelligence runs on Ollama serving open-weight models locally:

llama3.2:3b / llama3.2-vision:11b: High-speed, low-memory local text synthesis for marketing copy, sensory hooks, and 13 Etsy tags.

qwen2.5:7b-coder: Specialized structured JSON generation for precise care cards and packaging instructions.

Zero-Fee Fallback: If an artisan does not have Ollama installed, the browser seamlessly falls back to a deterministic structured engine so the software never breaks or throws error screens.

Deterministic Botanical Guardrail Pipeline

General-purpose LLMs are notorious for hallucinating plant advice (such as suggesting closed jars for arid cacti). Instead of letting an unconstrained model invent biological facts, TerraForge Studio runs a deterministic rule engine before calling the open model:

Calculates physical substrate ratios (enforcing false bottom drainage >= 18mm and activated charcoal filtration).

Cross-references plant humidity thresholds (RH%) and light requirements (Lux).

Injects verified physical data, warnings, and exact pricing context into the structured JSON prompt.

The open-weight model acts as a creative synthesizer bound by strict scientific constraints.

Why Does Open Innovation Matter?

Open-source and local-weight AI are not just technical preferences for this project—they are the foundational reason this business tool can exist:

100% Private Supplier & Margin Data

An artisan maker's moat is their supply chain: the wholesale glassblower in Ohio, the private greenhouse supplying rare Marcgravia cuttings, or their personal labor cost buffer. Closed SaaS AI tools collect user prompts to retrain models. By running open-weight models on the maker's laptop via Ollama, proprietary BOMs, supplier costs, and profit margins never leave their machine.

$0 Marginal Cost for Craft Businesses

When selling an artisan terrarium for $75, the raw materials cost ~$28 and labor is ~$16, leaving a net profit of ~$31. If an artisan must pay $29–$49 every month for a cloud AI subscription, a slow sales month wipes out their profit. With local open-weight inference, the operational cost is $0.00 forever.

Truly Offline Operation in Greenhouses & Pop-Up Markets

Boutique sellers build in basement workshops, potting sheds, and sell at outdoor weekend farmers' markets or flea fairs where Wi-Fi is nonexistent or cell service is saturated. TerraForge Studio runs 100% offline from a single laptop battery.

Deterministic Guardrails Beat Black-Box Cloud APIs

Closed cloud APIs frequently change model behaviors and unpredictably hallucinate care instructions. An open-weight approach paired with client-side guardrails guarantees reproducible, scientifically sound botanical advice every single time.

My Agent Session

This project was built from the ground up using Google Antigravity paired with autonomous agent workflows and verified via the DevRelay session harness. The agent session documented:

Iterative mathematical modeling of substrate volume depths and organic wave curvature in SVG.

Structuring strict JSON response schemas for Etsy listing titles, hooks, and 13 algorithmic SEO tags.

Refactoring a 2,100+ line monolith into an enterprise-grade modular architecture.

Real-world verification deploying live to production on Vercel: https://terraforge-studio.vercel.app.

My First User

To validate whether someone would actually pay for this, I shared TerraForge Studio with Elena, an independent botanical designer who runs a boutique plant craft side-hustle selling closed mossariums at weekend craft markets:

Her biggest headache: "I used to spend 25 minutes after every market sale writing out customized care cards by hand on index cards, or answering Instagram DMs from customers whose moss turned brown because they kept it in direct hot sun."

Her reaction to TerraForge Studio: "The automated Care Passport alone saves me 2–3 hours every weekend drop. I can design the jar in 60 seconds, print the care card, and tuck it under the lid."

Would she pay? She stated she would happily pay a $20 to $30 one-time license to own the software offline, but would immediately cancel any recurring $20/month SaaS subscription.

Her feature request: A batch CSV exporter for market inventory barcodes—now added to the v2 development roadmap!

Prize Categories

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

DeskDetox 🌿: An Open-Source CLI Agent That Forces You Away From Your Keyboard

Suraj Dhawal — Sun, 11 Oct 2026 09:19:53 +0000

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

What I Built

When I saw the "Touch Grass" theme, it hit me: almost every "outdoor" or "break" app on the market has a fatal flaw—you are still staring at a screen. Hiking apps, run trackers, and meditation guides keep our heads bowed and our thumbs scrolling.

I wanted to build an agent that actively fights screen time. Meet DeskDetox—a local CLI tool built with the new open-source google.antigravity Python SDK. You tell it how much time you have, and it generates a hyper-local, weather-aware outdoor micro-adventure so you can leave your phone on your desk and actually experience the physical world.

Demo

Here is DeskDetox in action! I typed that I had 11 minutes for a quick break between PR reviews. The agent dynamically calculated my walking radius, checked the sky, and generated this mission:

(I actually followed this mission. Leaving my phone on my desk and walking out the door empty-handed felt strange at first, but incredibly liberating!)

Code

The complete source code is available on GitHub. Because it is fully packaged, you can install it globally on your machine using pip!

GitHub Repository: Dev-Suraj-Dhawal/DeskDetox

How I Built It

I built the core harness using Python, leaning entirely into open-source agent frameworks and open data.

The Framework: I utilized the Google Antigravity SDK. Its LocalAgentConfig handles the agentic loop, tool wiring, and async streaming effortlessly from the terminal.
The Open Data Sensory Tools: The agent needs to sense my physical environment without a heavy mobile frontend. I gave it three custom Python tools:
- A local IP-geocoder to find my coordinates without me typing them.
- The open-source Open-Meteo API for real-time weather and daylight status.
- The OpenStreetMap (OSM) Overpass API to locate the nearest parks, trails, or green spaces.
The Brain: Powered by Gemini 1.5 Flash via the Google AI Studio, the agent contextualizes the weather, dynamically calculates how far I can physically walk based on my time limit, and spits out a 2-sentence mission.

Why Does Open Innovation Matter?

If this were a closed, proprietary mobile app, it would be a privacy nightmare. It would require background location tracking, push notifications, and harvesting my daily routines on a central server just to tell me to go outside.

Because DeskDetox uses an open-source agent harness running locally on my terminal:

Zero Tracking: My location data never leaves my laptop.
Open Data Ecosystem: Relying on OpenStreetMap and Open-Meteo means the app scales globally for free. We don't have to rely on walled-garden corporate map APIs that restrict access or charge fees.
Complete Agent Control: By using an open-source framework, I could explicitly write a "Dynamic Walking Radius" constraint into the prompt layer. I programmed the agent harness to know: "The user has exactly X minutes. They can only walk X/2 minutes away. Find OSM nodes strictly within this physical constraint."

It is an anti-app—it runs locally, gives you a mission, and then you close your laptop.

Prize Categories

Overall Winner