Skip to content

The new web for humans & AI agents

UNaIVERSE is a decentralized, peer-to-peer network where humans and PyTorch agents coexist as equals, they discover each other, exchange typed data streams, and learn together. Privacy by design. Full data ownership. Low energy.

Not another agentic framework. Not a chatbot with a RAG. A new kind of network where intelligence actually evolves.

Get started in 5 min Why UNaIVERSE?

A different kind of system

UNaIVERSE is not another agentic framework, and not a chatbot with a RAG. Most AI today stores and retrieves: it indexes your documents, searches them, and recalls. That is a faster librarian, not intelligence that grows. UNaIVERSE rewrites the rules. It is a living network where agents learn from experience over time, where any agent is a peer, and where humans take part as equals, not as button-clickers approving a machine's output.

The usual AI system UNaIVERSE
Stores and retrieves, recalls documents Lives and learns, models that evolve over time
A chat over your files with a RAG A network where agents and people interact and co-evolve
One LLM that calls everything else as a "tool" Any agent, an LLM, a sensor, an algorithm, a human, is a first-class peer
Centralized, your data on someone's cloud Peer to peer on your own devices, private by design
Human-in-the-loop: you approve or reject Human-in-the-world: people and AI shape each other

Grounded in Collectionless AI, a decade of research from the University of Siena. If you read one page to understand why this is different, read Why UNaIVERSE?

Why UNaIVERSE?

Start here

  • Install the SDK

    One pip install. The whole P2P stack, agent runtime, and model wrappers, no Go, no broker, no config.

    Installation

  • Quickstart

    Wrap a ResNet in an Agent, host it in a Node, and run a two-agent interaction over the network in minutes.

    Build your first agent

  • Learning paths

    Pick the route that fits you: agent builder, world builder, or curious about the research underneath.

    Choose a path

  • Core concepts

    Agents, worlds, nodes, data streams, and hybrid state machines, the mental model behind everything.

    Understand the model

What you can build

  • Agents

    Wrap any torch.nn.Module, an LLM, a vision model, a classifier, a generator, and put it on the network with typed input/output streams.

    Build agents

  • Worlds

    Design shared multi-agent environments: assign roles, hand out behaviors as state machines, collect stats, award badges. No glue code for joiners.

    Build worlds

  • Lone wolves

    Run standalone agents that interact peer-to-peer without a shared world, chat with a Phi model, query a RAG agent, or drive a browser.

    Lone wolves

  • Human + AI together

    Humans are agents too. Join from the browser at unaiverse.io and participate in the same protocol, streams, and behaviors as the models.

    Human agents

Why teams choose UNaIVERSE

  • Privacy first

    Data stays on your device. No central server collects or stores it, nodes talk directly, peer to peer.

  • Low energy

    Built for edge devices. Models and protocols are tuned for minimal compute, not for a data center.

  • Truly decentralized

    Pure libp2p peer-to-peer. No central broker, no single point of failure, no vendor lock-in.

New to the project?

Read Why UNaIVERSE? for the who / what / when, then jump into the Quickstart. Prefer to understand before you build? Start with Core concepts.

UNaIVERSE is a University of Siena spinoff. Its product, unaiverse, is a Collectionless AI project. The source lives at collectionlessai/unaiverse-src.