The IVIA Middleware
Accelerating Human-AI Research & Development
Stop spending time on repetitive engineering tasks. Focus on what matters: developing novel algorithms, visualizations, and interaction paradigms for Human-AI collaboration systems.
The Problem:
Isolated Silos
Modern AI systems require multiple components to work together. But researchers waste time rebuilding the same infrastructure for every project:
Duplicated infrastructure: Every app rebuilds databases, model inference services, auth systems, ...
No reusability: Apps are huge monoliths where components can't be shared between projects
Brittle systems: Tightly coupled code makes changes and updates risky and time-consuming
One-off prototypes: Code is hard to maintain, share, or reproduce across different environments
App 1
App 2
App 3
Applications
Middleware Orchestration Layer
Intelligent routing • Service discovery • Load balancing
Reusable Services
The Vision:
Reusable Components
Instead of isolated silos, the IVIA Middleware enables a new paradigm: build once, use everywhere.
Pluggable Services: Mix and match model inference, databases, auth — like LEGO bricks
Orchestration Layer: Intelligent routing and coordination across all services
Shared Infrastructure: One service powers multiple applications — no duplication
Rapid Prototyping: From the idea to a working system in hours, not months
The IVIA Middleware streamlines the rapid development of interactive AI and data processing workflows. In the following, we will explore its architecture and explain how it helps researchers, developers, and end users to build and configure workflows.
The Backend You Don't Have to Build
Move from monolithic headaches and migration hell to
a scalable microservice architecture out of the box.
From Silos to Services
Modern human-AI collaboration apps are complex. They need GPU inference, databases, auth, and more. IVIA Middleware replaces this tangled mess with a clean, API-driven service layer.
Intelligent API Gateway
Our middleware acts as a smart API gateway, routing requests, managing state, and ensuring your services work together seamlessly. No more writing brittle integration code.
A Toolkit of Production-Ready APIs
Access a growing gallery of powerful human-AI collaboration and infrastructure services through standardized APIs. What used to take months to build is now a single API call away.
Example: How an Application Is Built on the Middleware
generAItor is one of many complex visual analytics applications. Its parts map onto four shared layers that any other application can reuse.
Based on the generAItor system, we demonstrate how a complex Visual Analytics application decomposes into the four layers of the Middleware.
A Growing Gallery of Services
Every service in the catalog is built once and registered with the Middleware. From then on, any application can reuse it.
Services on the Middleware as of October 2026. The live catalog is in the Middleware portal.
Research Extension
Introducing Bonsai:Agentic Co-Development on the Middleware
We extended the IVIA Middleware with a mixed-initiative workspace where human and AI agents collaboratively build Visual Analytics applications — enforcing architectural guardrails at every step.
Plan
Decompose tasks and manage agents. Augmented task management ensures modularity and reuse from the start.
Design
Human and AI agents collaborate within bounded contexts, enforcing interface contracts across all four Middleware layers.
Monitor
Comprehensive provenance tracking of all multi-agent interactions. Safe collaborative development with full audit trails.
Review
Agentic evaluation, automated testing, and performance dashboards validate the resulting VA application before deployment.
Preprint / arXiv: 2604.19247
Bonsai: A Mixed-Initiative Workspace for Human-AI Co-Development of Visual Analytics Applications
Thilo Spinner, Matthias Miller, Fabian Sperrle-Roth, and Mennatallah El-Assady · 2026
Human-AI Co-Development – Building on the IVIA Middleware (Layered Core)
See It In Action
Watch how researchers at ETH IVIA Lab create powerful pipelines in minutes, not months
Real-World Applications
These are complete systems built by researchers at ETH IVIA Lab using the middleware to orchestrate multiple services
Each application combines multiple pluggable services from the gallery into cohesive research tools
explAIner
Combines XAI services, visualization services, and data management to create an integrated platform for model understanding.
LingVis
Orchestrates NLP models, embedding services, and collaborative tools for comprehensive visual text analytics.
Visual Musicology
Combines audio processing, document parsing, and visualization services for musical analysis and understanding.
Ready to Accelerate Your Research?
Join researchers at ETH IVIA Lab and beyond. Stop reinventing the wheel and start building the future of Human-AI collaboration.