HiNet.

Documentation

The whole design, open to read.

From the whitepaper down to the network-layer specs — how a sovereign local model becomes a decentralized intelligence network.

Start here

WhitepaperWhat HiNet is, why sovereign local intelligence, and how iCores compose into something bigger. The vision — iCore · iQuorum · iCorpOne model at three scales: you, the quorum you join, and the organization you build. Fractal Intelligence (origin paper)The founding conceptual paper — quorum sensing and fractal, holographic intelligence.

The network layer

Network integration noteHow the pieces fit — the cross-cutting contracts and the phased P0→P3 build roadmap. Node identity & registryOwner-held Ed25519 identity, self-certifying node ids, and the signed registry that indexes them. Routing & quorum aggregationThe three ask-modes, the dynamic router, Quorum Memory, and how quorums assemble per question. The grand monorepoOne codebase on a decentralized filesystem — every contribution signed and attributable. Onboarding & agent harnessThe first-run funnel and the dev-agent harness that serves the owner of an iCore. Owner identity & proof-of-personhoodA portable, biometric-gated owner key that certifies each node; personhood as an external credential — never biometrics in-network. Query classification & quorum cacheThe routing hot path: query → classify → order-invariant topic-set → cache → quorum (same topics, any order → same quorum). Quorumz map integrationThe node pool as a layer on the Quorumz map — geographic and by-domain/expertise, with an opt-in coarse-location privacy gate. Node distribution, EVM identity & senseThe downloadable self-updating node, its owner + iCore EVM keys, and the iCore → sense.quorumz signed-geobit flow.

The model & learning

The foundational modelThe fractal-holographic anchor and the composable deltas that make quorum composition possible. Personalization engineHow your iCore learns from your data — and the hardware that runs it privately, on-device. RAG architectureRetrieval that grounds every answer in your own sources, with citations — the memory under the model.

Research & validation

E1 — holographic compositionThe experiment that tests the core claim: independently-trained iCores composing into something better. The sim-to-real gapAn honest accounting of how far the experiments are from the production system — and what closes it.

Delivery & packaging

Node packaging & compute backendHow a node is packaged — the Rust security core, the compute-backend seam, and the trust boundary.

Technical specs

Technical specThe buildable surface — runtime, RAG, the compute seam, and the network-layer contracts. Track 2 — personal dataBringing your personal data in: connectors, sensitivity, and on-device processing. Canonical record (HCR)The shared record format every iCore reads and writes — the canonical item and its provenance.

HiNet · i.quorumz.com · early research — a living, versioned design.