Distributed Architecture

Purpose

To provide a decentralized, permissionless execution layer dedicated to a single, community-governed AI model/service.

The system is designed to survive partial failures and operate without trusted coordinators.

Core Principles

No single point of failure
No central authority
Graceful degradation
Horizontal scalability
Single-service scope over general-purpose compute

Key Concepts

Peer-to-peer coordination
Task and model sharding
Redundant computation
Bandwidth-aware protocols
Incentive-neutral participation
Trust & validation for permissionless agents

See also

Repository

The implementation lives in compute-all: orchestrator, PNode runtime, CLI, tokenizer, and monitoring stack in a single Bazel workspace.

Status

Actively evolving — see Project Status. Expect breaking changes.