Architecture

Separate your application from model and provider choice.

Workflow logic, context and state stay stable while models, providers, routes and compute surfaces can change underneath.

System anatomy

A routing brain, not a model wrapper

The architecture separates task intent, provider selection, execution controls and feedback signals.

Router
Dispatcher
Memory

System boundary

The workflow is the constant. The route is a variable.

INTENT01ORCHESTRATION02MODELS · TOOLS03CONTROLLED OUTPUT
Stays stableWorkflow logicContextStateBusiness intent
Orchestration layerAttractorFactoryroute · state · memory · control
Execution space · AttractorFactory trajectories
q₁q₂t
01Model
02Provider
03Route
04Compute surface
Control planepoliciesbudgetsreview pointsexecution traces

Below the surface

Legible technical layers, not a black box.

Each level maps to the engine's actual vocabulary.

  1. 01
    IntakeTask intake and decomposition
  2. 02
    ConstraintsPolicy, budget, privacy and risk gates
  3. 03
    RoutingOpen-model fleet: local, hosted, freemium — premium optional
  4. 04
    ExecutionMulti-agent execution with review, test and retry loops
  5. 05
    FeedbackTelemetry, outcomes and convergence signals
  6. 06
    EvidenceReview points, execution traces and integrations

One observable execution, end to end