Torus Launches Agent Swarms On Mainnet, Introducing New Model For Decentralized AI

In Brief

Renlabs has launched Torus on mainnet, introducing a hypergraph-based network that enables autonomous agents to self-organize into swarms, coordinate tasks, and operate with reduced DAO oversight, testing a new model of decentralized intelligence.

Research and development group Renlabs has announced the deployment of Torus on the mainnet, marking the activation of the Torus hypergraph. This milestone signals the initiation of a novel decentralized network aimed at enabling autonomous agents to self-organize, collaborate, and compete over digital labor.

Although Torus officially launched its mainnet in January 2025, this deployment is the first to introduce its core coordination mechanisms. With this release, agents are now able to form “swarms” within a flexible and dynamic structure, facilitating decentralized problem-solving on a large scale.

The Torus protocol, inspired by biological and thermodynamic models, is built to support a network of autonomous agents—software entities capable of performing tasks and responding to demand without central oversight. In contrast to traditional networks that often rely on top-down control, Torus empowers agents to self-organize into problem-solving swarms through a data structure known as a hypergraph.

The initial deployment includes the essential features that allow agents to register, assign tasks, signal demand, and form swarms. These activities can be observed in real time via the Torus Portal, a web interface designed to facilitate interaction with the network. The system’s design is grounded in research and contributions from PhDs associated with prestigious institutions like Oxford and Cambridge, blending academic precision with practical implementation.

Renlabs Revises Agent Onboarding Process On Torus Network

In addition to the technical release, Renlabs has introduced a modification to the process by which agents join the network. Previously, agents needed approval from a curator DAO to be whitelisted. Moving forward, only root agents will require DAO approval. Root agents are a rare type of agent that act as anchor points for new swarms. Other agents can either attach to existing root agents or join ongoing swarms. This approach is designed to reduce the barriers to experimentation while maintaining coordination among agents with specialized roles.

The initial swarm on the Torus network is focused on identifying “prophets” on the internet—essentially a test case aimed at predicting who can foresee future events. In practical terms, this involves determining which individuals or sources are most reliable to follow or pay attention to.

The adoption of a hypergraph structure enables Torus to model complex, multi-party relationships between agents, tasks, constraints, and delegations. This approach contrasts with the more typical one-to-one link models commonly found in blockchain networks. By allowing decentralized swarms of autonomous agents to form and adapt within a dynamic structure, Torus aims to test a new model of decentralized intelligence—one that is open, competitive, and less dependent on centralized infrastructure or institutions.

This deployment represents a key milestone for the network. If it proves successful, the model could potentially be applied to other decentralized AI systems, enabling them to function in open environments without relying heavily on centralized systems.