Skip to Content
Newnadoon AI FrameworkOverview

Newnadoon AI Framework Overview

Build, extend, and operate AI systems that automate real company operations using the Newnadoon TypeScript framework.

This documentation is written for engineering teams and operations stakeholders building production automation together.

Newnadoon is designed around a practical idea: business work is not one prompt. It is coordinated execution across teams, tools, systems, and memory over time.

Mission

Newnadoon helps teams automate business processes, workflows, and recurring tasks with production-minded agent architecture.

Instead of treating agents as isolated chat loops, Newnadoon models how companies operate:

  • Staff agents execute specialized tasks.
  • Department agents supervise and delegate.
  • Workflows orchestrate execution in defined sequences (with conditional routing and parallel branches when needed).
  • Tools connect agents to real systems (APIs, data services, internal platforms, MCP tools).

Core Benefits

  • Company-first architecture: model org-like behavior with manager/supervisor and worker patterns.
  • Stateful execution: preserve run context, checkpoints, and agent memory across steps.
  • Provider flexibility: use OpenAI, Ollama, or custom LLM providers behind a consistent interface.
  • Tool-native automation: support structured tool calling and integration with external systems.
  • MCP interoperability (bidirectional): call MCP servers from agents and expose MCP endpoints for external clients.
  • Debug and observability: stream events and inspect execution through debug tooling.
  • Voice agents: add realtime speech with NewnaVoiceKit (createVoiceAgent) using the same workflows as text chat.
  • Production surfaces: ship with API/server utilities and CLI entry points for running workflows in real environments.
  • Cloud hosting: deploy agents to Newnadoon infra with npx newnadoon deploy (public App, API, and MCP URLs).

Framework Model

1. Staff Agents (Execution Layer)

Staff agents are task executors. They reason, call tools, observe results, and return outputs for downstream steps.

2. Department Agents (Supervision Layer)

Department agents act like supervisors. They route work to the right staff agents and coordinate multi-agent completion.

3. Workflows (Orchestration Layer)

Workflows provide deterministic orchestration for end-to-end business processes with explicit transitions, sequencing, and state management.

What You Can Build

  • Internal operations agents (support, finance ops, logistics, compliance workflows)
  • Multi-step process automation with human-review points
  • Agent systems that combine LLM reasoning with API and MCP tool execution
  • Stateful agent services exposed through API endpoints

Install

npm install @newnadoon/ai-framework
Docs HomeGet started