Overview
A subnet in Konnex is a workload class: a real commercial pattern (aerial inspection, manipulation, 3D mapping) with a defined input schema, scoring function, and open miner competition. The design follows the same coordination shape as Bittensor-style subnets, while Konnex specializes verification for physical evidence, not only statistical text quality.
Testnet: three live subnets
Drone navigation
Vision–language flight; scoring on task match, safety, and execution
Roboarm VLA
Text-directed VLA manipulation (bi-manual, kitchen-grade examples); safety, task match, efficiency
SLAM 3D map
Sensor bundles; mesh / semantic map quality vs ground truth
Dashboard URLs for each workload will be published in official release notes when they are finalized. Get started: testnet quest flow and release notes.
Signed enterprise counterparties participate in these workload classes; full commercial runs may still live in private environments while the public chain proves verification and the roadmap advances mainnet settlement. The testnet is the place to try the Proof-of-Physical-Work loop end to end.
How a subnet is structured (conceptual)
Task schema and prompt format
KPIs and verifier checks (simulation, replay, sensor signatures, held-out data)
Scoring weights and reward split (testKNX on testnet; design target includes stablecoin settlement on mainnet per the roadmap)
APIs and binaries for miners and validators (see the SDK and konnex-network on GitHub)
Pages in this section
Related reading
Last updated