OpenCog Hyperon

The open AGI stack

Hyperon is SingularityNET’s open AGI stack: a single platform on which symbolic reasoning, neural learning, and probabilistic inference operate together at scale. Where most current systems scale a single approach, Hyperon integrates several within one architecture designed to reorganise and improve itself over time. It is the foundation on which the rest of this stack is built.

The stack consists of several core components, each of which can be learned independently. Their relationship is as follows:

• Hyperon is the framework.
• AtomSpace is the knowledge store within it, with DAS its scalable, distributed form.
• MeTTa is the language written for it.
• PLN and NARS are reasoning engines that run on it.
• OmegaClaw is an agent built on top, and the recommended starting point.

Those new to the stack should begin with OmegaClaw, run an agent, and return to these components to understand or extend its behaviour. The full vision and reference are maintained at hyperon.dev. Explore Hyperon →

MeTTa — the language of Hyperon

MeTTa (Meta Type Talk) is the language written for Hyperon. It combines functional, logical, and process-calculus styles, and a MeTTa program executes by querying and rewriting the knowledge graph directly. It supports constructs that most languages handle poorly, including code that modifies itself at runtime. Its model differs substantially from conventional languages, so the tutorials are best worked through in sequence. Learn MeTTa at metta-lang.dev →

AtomSpace — the knowledge store

The AtomSpace is Hyperon’s knowledge repository: a weighted, labelled graph that holds everything the system represents, from facts and rules to goals and procedures, within a single structure. MeTTa programs operate by reading and rewriting portions of it. The Distributed AtomSpace (DAS) extends this store across multiple machines, allowing a knowledge base to run in memory for speed and to scale across a cluster for size. Learn AtomSpace → coming soon

PLN — reasoning under uncertainty

Probabilistic Logic Networks (PLN) enable reasoning where evidence is incomplete or uncertain, as it is in most real conditions. PLN combines probability with logic, so that conclusions carry a measure of their likelihood rather than a binary true or false. It is one of the reasoning engines an agent such as OmegaClaw draws on. Learn PLN → coming soon

NARS — reasoning with limited knowledge and resources

The Non-Axiomatic Reasoning System (NARS), based on Non-Axiomatic Logic (NAL), is designed to reason under conditions of insufficient knowledge and limited resources. Rather than assuming complete information, it reasons from experience and revises its beliefs as new evidence arrives. It represents a different approach to machine reasoning from PLN, and Hyperon can draw on both. Learn NARS → coming soon

MeTTa-Motto — integrating language models

MeTTa-Motto connects large language models to MeTTa programs, allowing model generation and Hyperon’s symbolic reasoning to operate within a single workflow. It offers a practical entry point for combining neural and symbolic methods rather than selecting between them. Learn MeTTa-Motto → coming soon