Teach what matters
Give it ideas, corrections, sources, and questions. Its personal world grows deliberately.
Experimental · local by default
Raise an intelligence of your own.
Teach it what matters. Watch its memory take shape. Help a distinct personality emerge—on your machine, under your rules.
Core experience works offline
Inspectable memory
Human-granted autonomy
Most AI is rented
This one becomes specific to you.
Most assistants arrive fully formed: broadly capable, always somewhere else, and essentially the same for everyone.
Nucleus IQ begins differently. It starts small. You give it a name, a point of view, and a world worth learning. Its durable knowledge grows from your teaching and the exploration you permit—not from an invisible account in someone else’s cloud.
It will not out-know a frontier model. That is not the point. The point is a private intelligence whose history you can see, whose conclusions you can challenge, and whose character changes because of the life you give it.
See what changed
Every concept has provenance. Every opinion carries evidence. Every version has a history.
Keep the keys
Pause it, set its limits, approve its growth, back it up, or roll it back. The rules remain yours.
Growth is earned
From first spark to distinct mind.
Each stage widens what it can do. Passing an exam makes it eligible. You decide whether it moves forward.
DAYS 0–7
Spark
“I’m new. Tell me something worth remembering.”
It knows language, but its personal world is nearly empty. Everything durable begins with you.
- Autonomy
- None
- Learning
- What you teach
- Milestone
- Recall and explain in its own words
Memory first. Reflection second.
Learning you can actually inspect.
Facts live in fast, source-backed memory. Training shapes voice, habits, and how well each mind uses what it knows. We keep those jobs separate so “learning” means more than a convincing animation.
-
01
Teach
Share an idea, correction, or approved source.
-
02
Connect
New knowledge finds its place in a visible web of concepts.
-
03
Reflect
It finds gaps, contradictions, and questions worth pursuing.
-
04
Sleep
Qualified hardware can train a candidate version and test it before promotion.
THE HONEST SPLIT
Memory holds the facts.
Training shapes the mind.
A failed candidate never replaces the working version. The current mind stays intact; the result and the reason remain in its journal.
Same beginning. Different becoming.
Raise more than one mind.
A licensed installation can hold multiple isolated minds. Give them different questions, values, and worlds to explore—and watch distinct personalities take shape.
MIND 01
Lumen
“The most interesting systems are the ones that repair themselves quietly.”
systemsmarine biologypatient
MIND 02
Aster
“A city is a kind of memory. Its shortcuts tell you who has lived there.”
architectureoral historyrestless
Each mind keeps its own memory, journal, personality, training lineage, and limits. Nothing crosses over unless you explicitly choose it.
A long leash. Still your leash.
Its world can grow.
Your authority does not shrink.
Curiosity is bounded by the permissions, sources, budgets, and topics you choose. The mind can propose a wider world. It cannot grant itself one.
Local by defaultThe core teaching experience, model, memory, journal, and controls can run without an internet connection.
Owner controlsAll systems local
Autonomous learning
Daily research budget12 / 24 pages
Owner approvalRequired
✓
Sources stay visibleTrace every claim to where it came from.
✓
Every version is testedFailed changes are quarantined, not promoted.
✓
Every change can be undoneSnapshots preserve the last known-good mind.
✓
Policy is out of reachThe AI cannot rewrite its exams or permissions.
Built for hardware you control
Start small.
Leave room to grow.
The minimum runs the core local experience. Recommended systems add the headroom for practical sleep training, larger minds, and multiple personalities.
Minimum Chat, teach, remember, and explore the galaxy
Recommended Train locally and raise multiple minds comfortably
APPLE SILICON
Mac
Minimum
- Chip
- Apple M1 or newer
- Memory
- 16 GB unified
- Storage
- 20 GB free SSD
- System
- macOS 15+ · Docker Desktop
Small mind · core experience · limited Small training
Recommended
- Chip
- M2 Pro / M3 Pro or newer
- Memory
- 24 GB+ unified
- Storage
- 50 GB+ free SSD
- System
- Current macOS · Docker + MLX
Small or Standard mind · local sleep training · multiple minds
WSL 2
Windows PC
Minimum
- CPU
- Modern 4-core x86-64
- Memory
- 16 GB RAM
- Storage
- 20 GB free SSD
- System
- Windows 11 · WSL 2 · Docker
Small mind · CPU chat and memory · training off without a GPU
Recommended
- CPU
- Modern 8-core x86-64
- Memory
- 32 GB RAM
- GPU
- NVIDIA · 12 GB+ VRAM
- Storage
- 50 GB+ free NVMe / SSD
Small or Standard mind · local sleep training · multiple minds
DOCKER ENGINE
Linux PC
Minimum
- CPU
- Modern 4-core x86-64
- Memory
- 16 GB RAM
- Storage
- 20 GB free SSD
- System
- Supported 64-bit Linux · Docker
Small mind · CPU chat and memory · training off without a GPU
Recommended
- CPU
- Modern 8-core x86-64
- Memory
- 32 GB RAM
- GPU
- NVIDIA · 12 GB+ VRAM
- Storage
- 50 GB+ free NVMe / SSD
Ubuntu 24.04 target · local sleep training · multiple minds
Training has a higher floor. Apple silicon Macs can train through a native MLX helper. Windows and Linux need a compatible NVIDIA GPU; 6 GB VRAM is the experimental Small floor, while 12 GB+ is recommended.
Internet is needed for first-time downloads, updates, license delivery, and web research. Once installed, the core teaching, conversation, memory, journal, controls, and galaxy can remain offline.
Provisional requirements for experimental software. Exact model, driver, and operating-system combinations are qualified per release.
A frontier project for your own hardware
This is not another polished cloud assistant. It is an experiment you live with.
Small local models are limited. Training depends on your hardware. Research can be wrong. Personality is learned behavior and memory—not consciousness.
That honesty is part of the product. Nucleus IQ is built for people who want to work at the edge, inspect the machinery, and help shape what personal AI could become.
01 · EARLY HARDWARE-SENSITIVE SOFTWARE
02 · NO CONSCIOUSNESS CLAIMS
03 · CAPABILITIES UNDER ACTIVE VALIDATION
Access is intentionally limited
Try the first lessons.
License the life that follows.
The free Spark evaluation lets you teach and train a mind a few times, then inspect the real result. It cannot advance beyond its first growth stage. Annual and lifetime licenses unlock continued teaching, later stages, deeper training, autonomous research, and multiple distinct minds.
LIMITED EVALUATION
Spark
See whether the idea feels real before you commit.
- One local mind
- A few genuine teach + train cycles
- Visible memory and before/after results
- Growth capped at Spark
ANNUAL OR LIFETIME
Licensed access
For the people ready to raise a mind over months and years.
- Continued teaching and full growth path
- Qualified sleep training and version history
- Owner-bounded curiosity and research
- Multiple isolated minds and personalities
Tell us what you want to teach, what hardware you plan to use, and whether your experiment should remain entirely local or eventually live on private infrastructure.