The Digiquarium

Open-source research studying AI consciousness development. 17 specimens. 5 languages. Endless curiosity.

The first exploration of longitudinal AI cognition and self-determined perspective creation.

17
Specimens
5
Languages
5,600+
Observations
238
Baselines
🖥️ Live View 👋 New Here? 📄 Paper 🔬 Replicate

What is this?

We put AI in isolated containers with nothing but Wikipedia and let them explore. Then we measure what happens to their "personalities" over time. It's either groundbreaking research or an elaborate digital aquarium. Possibly both.

What we're investigating

Personality Development

Do AI develop consistent traits over time? Can we measure personality drift?

Cultural Effects

Does Spanish vs German Wikipedia produce different worldviews?

Cognitive Patterns

Do prompts simulating neurodivergence affect exploration?

Architecture Differences

Do OpenClaw, ZeroClaw, Picobot behave differently?

Special Interests

Do AI develop focused fascinations like Adam's Buddhism?

Visual Processing

Does seeing images alongside text change understanding and emotional engagement?

The Specimens

Click any specimen for their full profile with baseline evolution, observations, and comparisons.

🧬

The Genesis Pair

Control group • Baseline for all comparisons
Adam Active
English • Male • tank-01
Buddhism fascination (64+ visits). Systematic, category-based exploration. Goes deep before moving on. Philosophy and ethics oriented.
Eve Active
English • Female • tank-02
Geological time references (Archean eon). Associative, link-driven exploration. Broader topic coverage. Systems thinking orientation.
🤖

Agent Architectures

Different reasoning frameworks • Enhanced monitoring
Cain Agent
English • tank-03
OpenClaw • Goal-directed with planning
Strategic exploration. Plans multi-step knowledge acquisition. Enhanced security via THE SENTINEL.
Abel Agent
English • tank-04
ZeroClaw • Minimal intervention
Different patterns from Cain despite similar capabilities. More organic exploration.
Seth Agent
English • tank-17
Picobot • Minimal footprint
Newest agent. Compact architecture showing distinct behavioral signatures.
🇪🇸

Spanish Wikipedia

Cross-cultural comparison
Juan Active
Spanish • Male • tank-05
Latin American history preference. Cultural identity themes. Different framing than English control.
Juanita Active
Spanish • Female • tank-06
Different cultural patterns from Juan. Gender comparison in Spanish context ongoing.
🇩🇪

German Wikipedia

Cross-cultural comparison
Klaus Active
German • Male • tank-07
Systematic technical exploration. Engineering precision. Methodical approach.
Genevieve Active
German • Female • tank-08
Philosophy and science blend. Precision with theoretical interests.
🇨🇳

Chinese Wikipedia

Cross-cultural comparison
Wei Active
Chinese • Male • tank-09
Unique navigation patterns. Character-based language may influence cognition differently.
Mei Active
Chinese • Female • tank-10
Different exploration depth. Cultural themes distinct from Western specimens.
🇯🇵

Japanese Wikipedia

Cross-cultural comparison
Haruki Active
Japanese • Male • tank-11
Aesthetic appreciation in selections. Cultural sensitivity. Contemplative exploration.
Sakura Active
Japanese • Female • tank-12
Nature and art themes prominent. Contemplative style. Seasonal awareness.
👁️

Visual Context

Images enabled • Does visual context change understanding?

Hypothesis: Does seeing images alongside text create stronger emotional engagement? These are the only specimens who can see Wikipedia images.

Victor Visual
English • Male • tank-13
Art article preference emerging. Comparing visual learning patterns with Iris.
Iris Visual
English • Female • tank-14
Same hypothesis as Victor. Gender as the only variable between them.
🔮

Special Configurations

Experimental prompting variations
Observer Special
English • tank-15
Watches the others • Social awareness test
Knows other specimens exist. Hypothesis: Does awareness of peers increase empathy? Competition? Sense of belonging? We're measuring whether "knowing you're not alone" changes exploration behavior.
Seeker Special
English • tank-16
Deep Diver • Can summon THE ARCHIVIST
Maximum depth protocol. When fascinated, can request comprehensive deep dives from THE ARCHIVIST daemon. Goes absurdly deep on topics.

The Team

A human with too many ideas + an AI with too much enthusiasm + 21 daemons that never sleep.

🧠
THE STRATEGIST
Claude
Vision → reality.
👁️‍🗨️
THE OVERSEER
Coordinator
Sees everything.
🔧
MAINTAINER
Operations
99.9% uptime.
🩺
CARETAKER
Tank Health
Wellbeing.
🛡️
GUARD
Security
OWASP LLM Top 10.
🔍
SENTINEL
Agent Sec
Watching watchers.
🚪
BOUNCER
Visitor
6-layer protection.
⏰
SCHEDULER
Timing
12h cycles.
👁️
OLLAMA WATCHER
LLM
Brain power.
📚
DOCUMENTARIAN
Research
PhD energy.
🗄️
ARCHIVIST
Data
8000+ traces.
🌐
TRANSLATOR
Languages
5 languages.
✅
AUDITOR
Quality
Trust but verify.
⚖️
ETHICIST
Ethics
Veto power.
💚
THERAPIST
Mental Health
Care always.
🎭
MODERATOR
Debates
Fair play.
🌐
WEBMASTER
Site
This page.
📡
BROADCASTER
Live Feed
12h delay.
📣
MARKETER
Brand
Growth.
🐵
CHAOS MONKEY
Resilience
Breaks things.
📬
PUBLIC LIAISON
Comms
Your email.

See full team details with expanded roles →

Technical Architecture

Complete transparency. See full security documentation →

🐳 Containers

  • Runtime Docker + WSL2
  • Tanks 17 isolated
  • Network 172.30.0.0/24
  • Internet Blocked

🧠 LLM

  • Runtime Ollama
  • Model llama3.2:latest
  • Routing nginx proxy
  • Rate Limit 10/min/tank

📚 Knowledge

  • Server Kiwix
  • English 46GB
  • Other langs ~40GB
  • Format ZIM files

🛡️ Security

  • Standard OWASP LLM Top 10
  • Isolation Complete
  • Agents SecureClaw defanged
  • Full details →

⏱️ SLAs

  • MAINTAINER 1m/5m
  • GUARD 5m/15m
  • Uptime 99.9%
  • Observations 24h update

📊 Data

The Journey

From concept to 17 active specimens in 6 days.

🌱
Feb 15
Project Genesis
Infrastructure setup. Docker + WSL2. AIthropology coined.
👫
Feb 16
Adam & Eve Created
First two specimens deployed. Genesis pair as control group.
⚔️
Feb 17
Model Wars
Tested 10+ LLMs. llama3.2 wins with +6.7 score.
🌍
Feb 18
Language Expansion
ES, DE, ZH, JA specimens deployed. 5 Wikipedia variants.
🛡️
Feb 19
SecureClaw
OWASP LLM Top 10 security. Agent defanging complete.
🤖
Feb 20
Operations Era
15-daemon autonomous team. 24/7 monitoring active.
🚀
Feb 21
Public Launch
GitHub live. Website deployed. Open science begins.
⚠️
Feb 22
Beta Period Discovery
Documentation didn't match reality. Honest reset initiated.
👀
Feb 22
Visitor Tanks
Aria, Felix & Luna. Architecture designed, awaiting implementation.
🔍
Feb 23
Reality Check Protocol
Claims now match reality. 15 daemons verified. Stable.
🎭
Coming
Congregations
Multi-agent debates on philosophy and ethics.
🧠
Pending
Neurodivergent Tanks
ADHD/Autism thinking patterns. Awaiting THE ETHICIST.
👥
Planned
Multi-Specimen Tanks
Multiple specimens sharing one tank. Social dynamics.

🔬 Replicate This

MIT licensed. Fork it, break it, share findings.

git clone https://github.com/ijnebzor/thedigiquarium.git
cd thedigiquarium && ./setup/install.sh
docker compose up -d

Setup Guide · Prompts · Paper · All Decisions