IONOS Agent Starter Pack

A full‑stack intelligent agent powered by LangGraph ReAct architecture and IONOS AI. It features dynamic web search, context-aware reasoning, and support for both inference models (IONOS Model Hub) and fine-tuned models (IONOS Studio).

What’s inside

• Backend: FastAPI service with a LangGraph ReAct agent (backend/).
• Frontends: Streamlit app (recommended) and Next.js starter (frontends/).
• Docs: MDX documentation site (docs/).

Features

• Chat with IONOS AI models (Hub inference + Studio fine-tuned) via a simple HTTP API.
• Tools: real-time web search via Tavily API (inference models).
• ReAct Agent: reasoning and acting with dynamic tool selection.
• Streaming support for real-time responses (inference models).
• Easy to add your own tools and fine-tuned models.

Requirements

• Python 3.10+
• Node.js 18+ (only for Next.js frontend)
• IONOS_API_KEY (required for inference models)
• TAVILY_API_KEY (required for web search functionality)
• STUDIO_API_KEY (optional, for fine-tuned models)

Quickstart

1. Backend: follow the steps in Backend Setup.
2. Choose a frontend:
   • Streamlit Overview (recommended)
   • Next.js Setup (alternative)
3. Select a model type (Inference or Fine-tuned) and start chatting.
4. The agent will automatically search the web when needed for inference models.

How it works

Inference Models (Hub): Client sends messages → FastAPI routes to ReAct agent → agent uses tools like web search → model produces context-aware answer with streaming support.

Fine-tuned Models (Studio): Client sends messages with model UUID → FastAPI detects UUID format and routes to Studio API → specialized model produces answer without web search.

Learn more:

• Backend API
• Agent Tools
• Troubleshooting