Kvasir wiki

  • Node relay — A public address held on behalf of a machine that has none, so a node behind NAT can be reached without opening a single port.
  • Kvasir network — A decentralized AI-inference network (DePIN) where everyday devices serve open models and earn KVR.
  • Kvasir architecture — One map of the whole system: wallets, the gateway that takes payment, the bridge that fronts the engine, and the p4 network that runs the model.
  • Bridge — The HTTP face of the inference engine: what is loaded, who contributed, completions — and the market a contributing device joins.
  • Gateway — The public entry point: OpenAI/Anthropic-compatible APIs and KVR pay-per-inference settlement.
  • Node — Any device serving a share of a model — GPU, CPU, NPU or phone — earning KVR for the work it does.
  • p4 — The engine behind Kvasir: an event-addressed protocol where agents own nodes, stage servers hold layers, and placement is something an operator states rather than something the network guesses.
  • In-flight ring — A pipeline that never drains: several requests occupy different stages at the same time, so no stage waits for the one in front to finish.
  • Layer window & partial shards — A node's contiguous slice of the model — downloadable as a mini-GGUF instead of the full checkpoint.
  • Mixture of Experts (MoE) — A model whose FFNs are hundreds of independent experts, only a few of which fire per token.
  • Expert sharding — Splitting a MoE at the expert grain, so a phone carries 42–340 MB of experts instead of a 1.4 GB layer.
  • Router authority — The swarm's coherence invariant: routing is decided once, on the backbone — workers only receive expert ids.
  • Numerical equivalence — Different backends never agree bit-for-bit; the swarm treats measured tolerance as a first-class contract.
  • GGUF — The quantized model file format inference engine uses — and the layout that makes partial and expert slicing cheap.
  • KVR — The network's utility token: developers spend it on inference, contributors earn it for compute.
  • Contribution units — The reward formula: units follow tokens served weighted by layer share, then scale by performance tier.
  • Performance tiers — Measured throughput sets a multiplier: S ×1.5 · A ×1.25 · B ×1.0 · C ×0.7.
  • Staking — Locking KVR in the vault. It no longer gates operator roles, and nothing requires it.
  • Non-custodial wallet — Keys live only on the user's device — web, desktop, iOS and Android — and rewards settle straight to it.
  • SIWS + 2FA — Operator login is a wallet signature (Sign-In With Solana) over a server nonce, plus optional TOTP 2FA.
  • The KVR economy — How consumer cost and node reward form one self-reinforcing loop — the virtuous cycle that lets the network grow cheaper as it grows larger.
  • Inference pricing — What an inference costs in KVR today, why a decentralized network is structurally cheaper, and how price is meant to fall as supply grows.
  • Run an expert worker — Turn a spare GPU, CPU or phone into an expert worker: build it, volunteer for the scarcest slice, download it, serve, and dial out over 443 to earn KVR.
  • Operating a bridge — Operator notes for running the bridge and the ring behind it: load a plan, survive a restart, keep contribution flowing, and keep the engine off the internet.
  • WAN interconnect (200G optics) — How compute sites link at 200 Gb/s across a room, a campus, or a city: which optic at which distance, what plugs where, and what it takes to actually hit line rate.
  • Load-adaptive scaling — Kvasir's MoE serving path grows and shrinks with traffic: the coordinator re-engages proven workers under saturation, and the bridge recruits idle nodes by raising expert demand — all pull-based, so NAT'd devices join too.