Technology
Intelligence that remembers, reasons, and keeps improving.
Most large language models coming out of the big labs only get better when a new version is released. We’re building toward beneficial general intelligence that keeps what it learns, thinks through problems step by step, and improves while it works, so it becomes more useful the longer it runs.
Explore Hyperon01 / Approach
Our approach.
We’re building toward AGI the way minds work: memory, reasoning, and language working together in one system, with today’s large language models plugged in as parts. That system is Hyperon, and it’s the foundation for everything we build.
02 / Architecture
How it’s built.
Three core layers, with today’s large language models plugged in and decentralized hosting underneath:
Plugged in
Large language models
For language and perception. Every new model makes us stronger.
Three core layers
Core layer 01
Shared memory
What it remembers. One knowledge graph that scales.
Core layer 02
A language it can rewrite
How it’s programmed. Code the system can read and improve.
Core layer 03
Many kinds of reasoning
How it thinks. Logic, learning, evolution, and focus, working as one.
How it learns
Predictive and causal coding
How its neural networks learn. Hyperon is neural-symbolic, so it has its own approach to neural networks too. Training them with predictive and causal coding addresses problems that hold back standard neural networks, including continual, transfer, and lifelong learning. Together with neural-symbolic transformers and evolutionary self-organization, these methods are being built to work at modern AI scale.
Underneath
Decentralized hosting
Where it runs, so no single company controls it.
03 / Product
Omega.
Omega is how businesses put Hyperon to work, through AI agents that remember what they learn, reason through problems, and act within the rules they set.
04 / Safety
Beneficial and safe by design.
Beneficial means intelligence that gives people more capacity to shape their own lives. So safety is part of the architecture, not a policy on top of it.
Inspectable.
Every belief, goal, and decision can be read.
Tested before it changes.
Every self-improvement runs under test before it takes effect.
Approval before action.
Permissions, policy, and human sign-off sit between intention and consequence.
Decentralized.
No single power controls it.
Gated stages.
We don’t promise a date for AGI. Each stage must prove itself reliable, controllable, and auditable before the next begins, overseen by our Chief Ethics Officer and the board.
05 / Research
Research.
From OpenCog to Hyperon, our research community has spent 20 years building toward general intelligence: 750+ publications, 9,000+ citations, and an open core anyone can inspect.
20 years
From OpenCog to Hyperon
750+
Publications
9,000+
Citations