YonerAI

Platform architecture

Boundaries first, scale second.

YonerAI is split into client surfaces, run owner logic, tool workers, and storage so failures stay local and recoverable.

Client
Web / Mobile / Discord
Core
Run Owner Loop
Tool Plane
Skills + MCP + workers
State
Profile-separated storage

Core design rules

Single responsibility

Each layer has one job: UI, orchestration, execution, or persistence.

Deterministic run loop

Tool calls are tracked by IDs and results are deduped for idempotent behavior.

Upgrade without breakage

Public routes stay stable while internals evolve through versioned contracts.

Need transport details?

See how node pairing and multiplexed routing work in Relay model.