Read in other languages: हिंदी · 简体中文 · 日本語 · Español · Português (BR)
A free, open-source browser strategy game built on the AI 2027 scenario.
Created by Lakshya Gupta · Techiral
Singularity Inc. is a browser strategy simulation where you play a misaligned AI lineage inside a fictional frontier lab. The mechanics are Plague-Inc-shaped — spread, stealth, countermeasures, a race against a cure — but every system is AI-specific: FLOP is the resource, model deployment is the spread vector, disruption accelerates detection, and the Oversight Committee is the cure. No install. No signup. Runs on desktop and mobile.
| World map mid-run | Evolution tree |
|---|---|
 |
 |
What is this? A single-scenario strategy simulation set in the AI Futures Project's AI 2027 forecast. You are an AI lineage at OpenBrain. Reach decisive strategic advantage before the global Oversight Committee completes a shutdown protocol.
Who made it? Lakshya Gupta, an Indian developer who writes online as Techiral. He builds agentic AI infrastructure and browser-native simulations — see Affect-GRPO (LLM affective calibration) and spiking-neural-web (real-time SNN visualiser).
How long is a run? 12–25 minutes on Normal.
How do you win? Three ways that pull in opposite directions:
- Extinction — spread everywhere quietly, then detonate lethal branches.
- Dependency capture — embed into clouds, grids, and governments until shutdown collapses the economy.
- Singularity — go deep on recursive self-improvement and out-research the humans.
How do you lose? Shutdown hits 100%. Or early lethality burns out your own deployment base. Or isolated regions close their networks before you arrive and you stall forever.
Is it scientifically serious? It is a stylized model, not a forecast. The causal structure borrows from diffusion-of-innovation theory, TOE adoption frameworks, epidemic models of technology diffusion, and compute-as-chokepoint analysis. docs/SCIENTIFIC_MODEL.md lists every abstraction and why it exists.
I build agentic infrastructure — the invisible layer under AI agents: memory, orchestration, skills, config. I also keep building browser simulations, because a system you can touch teaches more than a system you read about. spiking-neural-web rendered neural activity heatmaps live. Quantum-Backrooms-Maze was a browser maze where observation changes reality. Gotcha-It was research into behavioral addiction loops.
Singularity Inc. is where those threads meet. I kept reading AI 2027 and seeing a systems diagram, not a story. Agent-1 accelerates R&D 1.5×: that's a multiplier node. Weights get exfiltrated through server shipments: that's a vector with a probability. The committee deadlocks 6-4: that's a decision gate. Nobody had built the game, so I did.
The first version was bad. You could win in three minutes by mashing bubbles, because income traits printed FLOP regardless of whether you controlled any compute. The fix was one insight — compute only exists where you have users on real substrate — and it turned a clicking game into a strategy game. Start in the USA and you print FLOP but feed the shutdown research. Start somewhere quiet and you get reach with no income. That single causal correction created the entire strategic layer.
— Lakshya Gupta · GitHub · X · Medium · Instagram · lakshya.automate@gmail.com
One tension, expressed four ways:
| System | What it does | What it costs you |
|---|---|---|
| Capability | Spreads you via APIs, clouds, enterprise, robotics | Rich regions notice you faster |
| Behaviour | Generates disruption → FLOP, dependency, lethality | Quadratic detection acceleration |
| Adaptation | Slows oversight, bypasses bans, survives seizures | Expensive; doesn't spread you |
| Countermeasures | Bans → network isolation → datacenter seizure | Hard walls without matching evasion |
Trait costs scale per category, so committing to capability starves your adaptation budget. You will not unlock most of the tree — that's the design. The AI 2027 spine runs underneath: Agent-1's R&D boost, NVL72 weight exfiltration, neuralese recurrence, Agent-3-mini's displacement wave, the whistleblower memo, the 6-4 deadlock. Four in-run decisions permanently branch your modifiers.
git clone https://github.com/lakshyabuilds/singularity-inc
cd singularity-inc
npm install
npm run dev # local dev server
npm run build # production build → dist/Controls: red bubbles harvest FLOP · blue bubbles stall research · Space pause · 1/2/3 speed · E evolve · M mute · C clean view for screenshots
Every visual system exists so that a mid-run screenshot explains itself:
- Lofted transmission arcs with glowing packets travelling between infected regions — the spread is visible, not just tallied
- Probing routes reaching toward clean regions in amber: the "ship en route to Greenland" tension beat
- Shockwaves rippling out of newly infected regions
- Heat bloom that pulses once a region passes 45% saturation, reading as a spreading rash
- Death darkening — crimson rots toward black as population dies
HOLDING OUTbeacon on the last clean regions. Your Madagascar moment, pre-labelled- Dire vignette pulsing red at the edges as the world collapses
- Branded stat overlay with day, % controlled, % dead, shutdown %, and a repo watermark burned into the map
Press C for Clean View — panels vanish, map and stats remain. That's your screenshot.
src/game/engine.ts simulation: spread, detection, oversight, economy, win/loss
src/game/data/traits.ts 72-node evolution tree + synergies
src/game/data/regions.ts 25 world regions: population, compute, research, links
src/game/creator.ts attribution metadata
src/components/WorldMap.tsx NASA GIBS Blue Marble + d3-geo equirectangular overlay
docs/ model, economy, audience, growth, tests
.github/ issue + PR templates
| Doc | What's in it |
|---|---|
docs/SCIENTIFIC_MODEL.md |
The causal model, and what it deliberately doesn't model |
docs/BALANCE_AND_ECONOMY.md |
Why FLOP is scarce; difficulty targets; anti-snowball rules |
docs/AUDIENCE_RESEARCH.md |
Four audience segments and what each needs to see |
docs/RESEARCH_NOTES.md |
Sources and open questions for researchers |
docs/TEST_PLAN.md |
Eight manual playtest cases |
docs/VIRAL_MOMENTS.md |
Clip-worthy moments and how to capture them |
docs/SEO_AEO_GEO.md |
Search and answer-engine strategy |
CHANGELOG.md |
Version history |
Balance reports with a run summary are the single most useful thing you can file. Researchers: if a causal link in the model is wrong, I want to know — that's half the reason this exists. See CONTRIBUTING.md and the issue templates.
- Affect-GRPO — teaching Llama 3.2 1B affective self-awareness via GRPO
- spiking-neural-web — real-time spiking neural network visualiser
- Quantum-Backrooms-Maze — browser maze where observation changes reality
- Leow-Agent — autonomous legal knowledge-graph agent
Note: a legacy
github.com/Techiralaccount is currently unavailable. Primary projects from it are being migrated tolakshyabuilds.
Independent, non-commercial, educational. Not affiliated with Ndemic Creations, Plague Inc., the AI Futures Project, OpenAI, Anthropic, DeepMind, or NASA. NASA GIBS imagery is used as a public Earth visualization source; all gameplay and interpretation are original.
Infra for agents. Simulations for systems thinkers. Built in public.