AttractorFactory

Architecture

A control layer for agents, models, budgets and review.

Work is separated from model choice. Each task is routed across local, open source, freemium or premium execution surfaces — through budget, privacy and review gates. High-level by design: the doctrine is public, the engine is not.

System anatomy

A routing brain, not a model wrapper

The architecture separates task intent, provider selection, execution, safety gates and feedback.

Router
Dispatcher
Memory

Operating principles

01

Router

Selects the execution route from the task's role, risk, context needs, budget and observed outcomes.
02

Dispatcher

Executes the chosen route, enforces budget and session gates, and records refusals and failures for inspection.
03

Memory

Keeps outcomes, routing decisions and feedback so the system learns where quality, speed and safety converge.

System layers

  1. 01Task intake and decomposition
  2. 02Policy, budget, privacy and risk gates
  3. 03Open-model fleet: local, hosted, freemium — premium optional
  4. 04Multi-agent execution with review, test and retry loops
  5. 05Telemetry, outcomes and convergence signals
  6. 06Human approval, audit trail and integrations