Open, Decentralized Artificial Intelligence
We believe that artificial intelligence should be a shared public resource, not a closed system controlled by a few.
Read more about why we are doing this.
This initiative exists to build a fully open-source AI ecosystem — from training algorithms to infrastructure to pre-trained models — developed transparently and owned by the community.
Importantly, our distributed network is designed to operate a single community-governed model and service. It is not a general-purpose public computing platform. Governance for training data, ethics, safeguards, and bias will be handled by a community process when the model reaches maturity.
Mission
Our mission is to create an AI stack where:
- Algorithms are open and auditable
- Pre-trained models are freely accessible
- Compute is distributed, voluntary, and decentralized — but narrowly scoped to serving and training a single open model/service
- Training is efficient without massive memory requirements
Every layer of the system is designed to be inspectable, reproducible, and community-owned.
Core Projects
🧠 Algorithms
Research into memory-efficient, decentralized training algorithms that work on heterogeneous machines and compose well with the distributed architecture.
🌐 Distributed Architecture
A permissionless, fault-tolerant compute network dedicated to a single open AI service. See the high-level Architecture Overview and dive into the Orchestrator, Compute Agent, Protocol, and Network Membership. Learn how we handle Trust & Validation and how to run Trusted Agents in the Cloud.
📦 Models
Fully open, reproducible pre-trained models built on top of the open infrastructure.
→ Models
Principles
- Open source is non-negotiable
- Transparency over convenience
- Decentralization over control
- Community ownership over profit
- Single-service scope over general-purpose compute
Roadmap
Our development follows a compute-first roadmap: we start by building a decentralized distributed computing network, then progressively use it for training and inference.
→ Roadmap
Learn the Architecture
- Overview: Architecture Overview
- Components: Compute Agent · Orchestrator · Protocol
- Network layer: Network Membership & Discovery
- Security: Threat Model
- Reliability & trust: Trust & Validation
Join the Movement
This is an evolving project.
The design is not finished — and that is intentional.