NexusNexus/ litepaper
nexus // litepaper v0.1

One robot learns. Every robot inherits.

Nexus is the open, model-agnostic skill layer for robots. A neutral protocol and registry where a learned skill routes onto any body, any brand.

Intro

The next decade of robotics will not be won by a single vendor. It will be won by the layer that lets every body learn from every other body. Nexus is that layer.

We sit on top of the open models. We do not replace π0, GR00T, or LeRobot. We make the skills they produce portable, routable, and rewarded.

The gap

The bottleneck to robot intelligence is no longer hardware or compute. It is data, locked in proprietary silos. LLMs trained on roughly 100,000 human-years of text. Robots have nowhere near the equivalent action data, and physical tasks need more data per skill than language.

The unlock is proven: DeepMind's Gemini Robotics 1.5 showed a skill trained on one arm just works on another body, no per-robot specialization.

But the response is walled gardens. Unitree's UniStore and OpenMind's robot app store both tie skills to their own hardware. Every fleet relearning to open a door from zero is a coordination failure paid for once per robot.

What Nexus is

  • Portable skill registry. Publish and consume learned skills, policies, and models across embodiments, not locked to one OS or maker.
  • Cross-embodiment routing. Translation across shared observation and action spaces, so a skill from one body maps onto another.
  • Tokenized incentive layer. Rewards for contributing and consuming skills. The flywheel fuel the open-model wave lacks.
  • Neutral protocol. Not a robot, not a model, not an OS. The layer that connects all of them.

Where Nexus fits

The open-model wave is real: π0 is open-sourced, NVIDIA ships GR00T, Open X-Embodiment spans over a million trajectories, LeRobot hit 58,000 datasets. LeRobot owns open data with no incentive layer. The crypto players sit in adjacent lanes (data, machine-economy, execution), none a neutral portable-skill registry.

Nexus is the tokenized, model-agnostic skill-routing layer on top of the open models, monetizing the transfer they already prove. Cross-embodiment training shows 50 to 200% success gains. That whitespace is ours.

The $NEXUS flywheel

Reframe the token as a royalty and inheritance economy, not a generic utility token.

Contribute
Publish a skill. Earn $NEXUS each time another body inherits it.
Route
Pay $NEXUS to route a skill onto your robot, across any brand or body.
Validate
Stake $NEXUS to attest that a skill actually transfers. Honest validation earns, bad-faith is slashed.
Govern
Holders steer routing standards and how the registry works.

$NEXUS is a utility token for the Nexus network. Nothing here is financial advice.

FAQ

Is the registry live today?
No. Cross-embodiment transfer is research-proven. The portable, tokenized registry is the thesis we deliver.
How is this different from LeRobot?
LeRobot is an open dataset and training library, and a great one. Nexus is a registry and routing layer with an incentive economy on top of open models like π0, GR00T, and LeRobot. We complement, not replace.
Why a token?
Because cross-embodiment value is a coordination problem. A royalty on inheritance pays the contributor each time their skill is reused. That is the missing flywheel.
Why launch on Virtuals (Base)?
Distribution, an agent-native community, and a launch surface that matches an open-protocol thesis.