RM2Lab AI Platform

Overview

The RM2Lab AI Platform provides access to locally hosted Large Language Models (LLMs) through both a user-friendly web interface and an OpenAI-compatible API.

The platform is built on:

  • Open WebUI – the primary web interface for interacting with AI models.
  • LiteLLM Proxy – an OpenAI-compatible API gateway exposing the available models for applications and developers.


Access

Access to the platform is available to all RM2Lab users.

Web Interface

The primary entry point is:

https://agent.rmlab.infn.it

Authentication is performed through INFN AAI.

Open WebUI

Open WebUI is the main interface for interacting with the hosted AI models through a modern web application.

Official documentation:

Open WebUI provides features such as:

  • Chat with multiple AI models
  • Conversation history
  • File uploads for model interaction
  • Custom system prompts
  • Model selection
  • Workspace organization
  • Personal knowledge and document interaction
  • Chat sharing
  • OpenAI-compatible API access

For a complete overview of the available features, please refer to the official Open WebUI documentation.

API Access

Chat Models

Every user can generate their own personal API keys directly from Open WebUI.

After logging into the platform, API keys can be created from the User Settings (top right) → Account → API Keys section of the personal dashboard.

These keys provide access to the available chat models through the platform’s OpenAI-compatible API and can be used with any compatible client or application.

Embedding and Reranking Models

Embedding and document reranking models are not currently available through Open WebUI API keys.

Users requiring these capabilities should request access to the LiteLLM Proxy platform.

LiteLLM exposes OpenAI-compatible endpoints suitable for:

  • text embeddings
  • semantic search
  • Retrieval-Augmented Generation (RAG)
  • document indexing
  • reranking

Access to these endpoints is granted upon request.

Available Models

The platform currently provides the following models:

Model

Purpose

Qwen3.6-27B

Dense 27B qwen model for coding / agentic / generic tasks

Qwen/Qwen3-Embedding-4B

Text embeddings for semantic search, RAG and document indexing

BAAI/bge-reranker-v2-m3

Document reranking for Retrieval-Augmented Generation (RAG) pipelines

Our goal is to continuously improve the quality of the hosted models as additional compute resources become available. New models may be added and existing ones may be upgraded or replaced over time.

Intended Usage

The platform is suitable for:

  • Software development
  • Code generation and review
  • Technical documentation
  • General-purpose AI assistance
  • Integration into applications using OpenAI-compatible APIs
  • Retrieval-Augmented Generation (RAG) workflows
  • Semantic search and document indexing


  • No labels