Architecture

AmarktAI Network is an AI orchestration infrastructure layer — not an assistant, not a wrapper.

It routes AI tasks through Aiva — the intelligent operations layer — which selects the right model, applies policies, handles fallback, and returns structured output. Built for operators running multiple AI products, not for one-off prompting.

Core architecture

Six layers that make the system work.

Aiva Intelligence

Natural language understanding, intelligent routing, and execution. Aiva interprets your intent, selects the right capability, and returns structured outputs.

Multi-Provider Routing

Adaptive model selection across providers with fallback logic, capability matching, and policy-controlled cost management.

Memory & Context

Cross-session retrieval and context persistence for apps that need intelligent continuity across requests.

App Agents

App-scoped agents with dedicated behavior constraints, capability policies, and per-app model selection.

Safety & Policy

Provider keys, capability restrictions, adult content controls, and execution policies built into the orchestration core.

Unified Workspace

Website, API, and admin console all map to one underlying operating model — one place to build, test, and deploy.

Execution flow

How a task moves through the system.

1

Describe your task

Tell Aiva what you need in natural language.

2

Aiva classifies

Routes to the best capability and model for the task.

3

Policy check

Capability permissions, content flags, fallback rules evaluated.

4

Execution

Model called with memory hooks and runtime observability.

5

Output + artifact

Response stored as artifact with provider/model metadata.

6

Operator trace

Events, usage, and artifacts visible in the operator console.

Smart routing first

Not a chatbot wrapper.

Aiva routes to the best model for each task. You are not locked to one provider. Policy, fallback, and cost logic are built in — not bolted on.

  • Multi-provider, not single-model dependent
  • Cost-tier model selection (best / balanced / cheap)
  • Automatic fallback when primary engine is unreachable

Operator control

Built for product operators.

Not a prompt playground. The operator console manages apps, agents, artifacts, GitHub repos, deployments, and provider configuration from one interface.

  • Multiple app surfaces under one routing brain
  • Workspace → GitHub → deploy in one flow
  • Artifacts, events, and traces in one view

For builders running production AI workloads.

Request access for teams that need orchestration quality, reliability, and real operational visibility.