Every claim is tied to a gate we can prove
We claim a capability only once its gate is crossed. There are no dated promises of general intelligence, and no superlatives that are not anchored to a measurable test.
Sovereign stack live
The strongest AI for your life. Private, and available where the others are blocked.
- +Personal AI ships inside the OrbVPN app on a top open model, self hosted
- +Compute network and node client for phones, PCs, and Macs
- +ORB unified across VPN, compute, and AI, with the DAO governance contracts live
- +Federated learning with differential privacy and secure aggregation
Wedge dominance
Provably first on your context. Blind preference wins against the frontier on personal tasks.
- +Continual training on the data flywheel plus reinforcement from real tasks
- +The continually trained model beats all open models in wedge domains
- +Blind A/B evaluation harness against frontier models on personal context
- +Inference verification live across the device fleet
Self improving and self releasing
An AI that improves itself, and ships its own upgrades, safely.
- +DAO owned GPU cluster plus staked operators, low communication distributed training
- +The champion challenger release pipeline runs in production
- +Automatic promotion under a safety council gate, canary rollout, auto rollback
- +Governance progressively transfers to the DAO on published milestones
General frontier program
Claimed only once crossed. The discipline that keeps every prior claim true.
- +Attempted only when treasury funded compute crosses an explicit threshold
- +Only on training recipes validated by the automated research loop
- +Pipeline parallel consumer GPU tier matures
- +Licensed distillation switched on if partnership agreements close
Out competing every centralized lab is the destination. The path is staged, and each gate funds the next. We publish the gate before we claim the capability, and the compute for the frontier fight is built from the treasury the earlier gates create.
Be part of Gate 0
The sovereign stack ships first. Run a node, or read the whitepaper for the full plan.