SwarmHarness: Decentralized AI Compute Network via Incentive-Aligned Protocol
SwarmHarness, a novel decentralized protocol, allows individuals with unused GPU cycles on personal computers, inference servers, and edge devices to securely and profitably share their computing power without relying on a central authority. Introduced in the arXiv paper 2605.28764, SwarmHarness features three key elements: a SwarmRegistry based on a Distributed Hash Table (DHT) for discovering peers and advertising capabilities; a SwarmRouter that allocates tasks based on a utility function considering capability, load, latency, and trust; and SwarmCredit, which incentivizes participating nodes with compute credits. Unlike existing solutions such as cloud marketplaces that depend on trusted coordinators, or Golem and BrokerChain's complex blockchain setups, SwarmHarness enables HarnessAPI skill nodes to autonomously form a compute swarm.
Key facts
- SwarmHarness is a decentralized protocol for sharing idle GPU compute resources.
- It targets personal workstations, idle inference servers, and edge devices.
- The protocol has three components: SwarmRegistry (DHT), SwarmRouter (utility-based task dispatch), and SwarmCredit (incentive mechanism).
- Existing solutions include cloud marketplaces, Golem, BrokerChain, BOINC, and Petals.
- SwarmHarness requires no central authority or heavy blockchain infrastructure.
- The paper is arXiv:2605.28764.
- HarnessAPI skill nodes self-organize into a compute swarm.
- SwarmCredit attributes compute-credit rewards to contributing nodes.
Entities
Institutions
- arXiv
- Golem
- BrokerChain
- BOINC
- Petals