KwaaiNet

P2P Infrastructure

A volunteer-operated P2P distributed AI inference network that shards large language model weights across participating nodes, enabling community-run AI computation without centralized cloud infrastructure. Built on the Petals framework (legacy Python implementation, OpenAI-Petal); being rewritten in Rust with WASM + WebGPU for browser and mobile deployment. Operated by Kwaai, a 501(c)3 nonprofit focused on democratizing personal AI through self-sovereign, privacy-preserving principles.

Comunidade

Detalhes

Licença CC-BY-4.0 (OpenAI-Petal); CC0-1.0 (repo license) https://github.com/Kwaai-AI-Lab/KwaaiNet
Status de Dev 🔬 Beta
Detalhe do Status de Dev Beta (legacy Python/Petals node operational; Rust rewrite in active development targeting Q1 2026 browser launch)
Proprietário Kwaai AI Lab (501c3 nonprofit)
País USA
Ano de Início 2023
Stack Python + Shell (legacy OpenAI-Petal); Rust + WASM + WebGPU (KwaaiNet rewrite in progress)
Financiamento Nonprofit (501c3); donations via Zeffy; undisclosed grants
Última Investigação 1 de jul. de 2026

Domínios de Caso de Uso

Platform Evolution & Sovereignty

P2P Infrastructure Atributos

Origins Distributed AI inference / Personal AI
Database N/A (inference routing infrastructure, no application data store)
Data Formats OpenAI API-compatible JSON (chat/completions endpoints); arbitrary LLM token streams
Mobile Support N/A (legacy); Planned — iOS/Android Q2–Q3 2026 (Rust rewrite)
Web Support N/A (legacy); Planned — browser extension Q1 2026 (Rust rewrite)
Native Apps Yes — Linux and macOS installers; Docker/Podman containers; CLI daemon management
Terms Free (volunteer node operation; no compensation to node operators)
Funds Undisclosed
Based On Petals (distributed inference framework, legacy); BigScience / Together AI Petals protocol
P2P Architecture DHT-based P2P mesh; nodes contribute model shards; bootstrap peers for initial discovery; OpenAI-compatible API gateway layer
Overlay Network Global (nodes distributed across public network; map.kwaai.ai provides node visibility)
Content Addressing No
Local-First N/A (inference routing, not storage)
E2EE No (data passes through participating nodes; node operators can observe inference traffic)
Byzantine Fault Tolerance No
Signature N/A
Permissions N/A (open relay; no cryptographic access control on inference requests in current implementation)
Protocol Maturity / Standardization Community Standard / Pre-specification (Petals protocol underlies legacy implementation; KwaaiNet Rust rewrite is pre-specification)
Infrastructure Function Distributed compute (AI inference)
Threat Model N/A (not designed as censorship circumvention infrastructure; primary threat is model weight confidentiality and inference privacy)
Volunteer / Node Model Volunteer-operated (community nodes); Foundation-operated (Kwaai operates bootstrap/coordination infrastructure)
Traffic Visibility Content visible to intermediate nodes (model shard holders observe token fragments during inference; E2EE not implemented)
Active Deployment Scale ~100 active nodes (legacy OpenAI-Petal network, developer/enthusiast community); KwaaiNet Rust rewrite pre-deployment