NLIP Swarm

NLIP Swarm is a University of Delaware senior design project built in collaboration with IBM to explore the Natural Language Interaction Protocol (NLIP) in a real-world sustainability scenario. Our team is creating a smartphone app paired with server-side AI agents that help farmers in Tanzania and Uganda access translation, lookup, and sensor-driven insights without needing proprietary vendor stacks.

Project Goal

Dinesh’s plan splits the work into two tracks:

• Mobile experience – a React Native (Expo) application that can use onboard sensors, collect farmer input, and securely exchange NLIP envelopes with the backend.
• Agent registrar & swarm manager – Python services that host and orchestrate NLIP-compliant agents (translation, text lookup, audio/image understanding today; more sophisticated chains later).

The long-term objective is to ship a reference implementation that demonstrates how vendor-neutral NLIP traffic can move between lightweight edge clients and richer cloud-based reasoning services.

NLIP in this project

NLIP is an open, multi-modal, JSON-based protocol for agent-to-agent communication. In NLIP Swarm we:

• Use NLIP envelopes to route requests between the Expo client and Python agents.
• Adhere to HTTPS bindings for transport, leaving room for future upgrades (WebRTC, QUIC, etc.).
• Focus on policy-aware, secure message exchange while keeping the rest of the stack cloud-agnostic.

Tech Stack

Frontend

• Framework: React Native with Expo
• Routing: Expo Router (file-based routing)
• Language: TypeScript

Backend

• Language: Python

Architecture Overview

frontend/   # Expo mobile client that captures text/audio/image inputs
backend/    # Python agent services, registrar, and NLIP session server
compose.yaml# Docker Compose for running both tiers together

Project Structure

nlip_swarm/
├── frontend/          # React Native mobile application
│   ├── app/          # Expo Router file-based routes
│   ├── components/   # Reusable UI components
│   ├── constants/    # Theme and app constants
│   └── hooks/        # Custom React hooks
└── backend/          # Python backend services

Getting Started

Frontend

1. Navigate to the frontend directory:

   cd frontend

2. Install dependencies:

   npm install

3. Start the development server:

   npm run start

4. Open the app:

   • Press i for iOS simulator
   • Press a for Android emulator
   • Press w for web browser
   • Scan QR code with Expo Go app on your device

Backend

1. Navigate to the backend directory:

   cd backend

2. Install Requirements

   pip install -r requirements.txt

3. Run The backend server

   python -m app.system.main

Docker Compose (full stack)

1. Create environment files from the variables below, or set them directly in your shell.

   Frontend — create frontend/.env:

   EXPO_PUBLIC_API_BASE=http://localhost:8024
   

   Backend — create backend/.env:

   DATABASE_URL=postgresql+asyncpg://postgres:postgres@db:5432/nlip_db
   NLIP_LOG_LEVEL=INFO
   # Optional: override LLM endpoint and model
   OLLAMA_URL=http://ollama:11434
   OLLAMA_MODEL=ai/llama3.2:3B-Q4_0
   

2. Start the full stack with Docker:

   docker compose up --build

3. The coordinator API will be available at http://localhost:8024, and Expo DevTools will be available on the forwarded ports.

4. If you are using Expo Go on a physical device, set EXPO_PUBLIC_API_BASE to your host LAN IP before running compose.

Contributors

• John Fulkerson jtfulky@udel.edu
• Mason Kulikowski masonkul@udel.edu
• Christopher Calderone ccald@udel.edu
• Kevin Kramer kkramer@udel.edu
• Benjamin Zlatin bzlatin@udel.edu
• Tyler Walsh tjwalsh@udel.edu