Data Systems Group

Max Planck Institute for Software Systems

Runtime guarantees for AI agents, and data systems for AI

About the Group

The Data Systems Group (DSG) at MPI-SWS builds systems for AI agents and data-intensive AI applications. We build prototypes, measure them on real workloads, and release code where we can.

Our current work covers runtime guarantees for AI agents, language models inside database systems, reproducible LLM pipelines, and large-scale graph learning. We collaborate with academic groups and industry partners in Europe and North America.

Learn more about our team

Laurent Bindschaedler

Research Group Leader

Runtime guarantees for AI agents, data systems for AI, and distributed systems.

Personal website

Research Areas

We work in four areas: agent runtimes, data systems for AI, trustworthy AI, and graph systems.

Agentic AI Infrastructure

Runtime contracts for AI agents: when an action becomes permanent, what state an agent keeps, and tamper-evident records of what it did.

Data Systems for AI

Language models as database operators, semantic caching for OLAP, and reproducible, lower-cost LLM pipelines.

Trustworthy AI

Measuring how context steers what models generate, and who decides what counts as correct when agents repair code.

Graph Systems

Gradient-only communication for large-scale GNN training, graph partitioning, and graph query evaluation.

View All Projects

Recent Publications

Our latest research

View all publications

ClawMMU: Protection-Coherent Virtual Memory for Stateful LLM Agents

2026

AgenticOS Workshop at NeurIPS 2026

Jinhao Hu, Bardia Mohammadi, Mofasshara Binte Rafique, Lars Henning Klein, Akhil Arora, Ashvin Goel, Laurent Bindschaedler

Agent Flight Recorder: Tamper-Evident Audit Trails with On-Chain Anchoring for Long-Horizon Tool-Using Agents

2026

BCCA 2026

Laurent Bindschaedler, Quentin Botha, Christoph Siebenbrunner

Bonded Recourse for Smart-Contract Settlement of Compensable Agent Side Effects

2026

BCCA 2026

Laurent Bindschaedler, Quentin Botha, Christoph Siebenbrunner

Ghost Tool Calls: Issue-Time Privacy for Speculative Agent Tools

2026

REALM 2026

Bardia Mohammadi, Lars Henning Klein, Akhil Arora, Laurent Bindschaedler

Latest News

Updates from the Data Systems Group

Publication

Four Non-Archival Workshop Papers on Agent Runtimes

Ghost Tool Calls, When to Cross the Rubicon, and The Working Set of a Coding Agent at REALM 2026, and ClawMMU at the AgenticOS Workshop at NeurIPS 2026.

October 2026

Publication

Two Papers Accepted at the ASE 2026 Workshops

ForkOps (SWGeno 2026) on operating fleets of generated software variants, and Whose Oracle Is It? (APR 2026) on authority in deployed agentic program repair, both at ASE 2026 in Munich.

September 2026

Publication

Three Papers Accepted at the SOSP 2026 Workshops

Externalization Barriers and The Irreversibility Budget at AgenticOS 2026 (both non-archival), and Don't Trust the Code, Check Its Effects at PACMI 2026. All three workshops are co-located with SOSP 2026 in Prague.

August 2026

Join Our Group

We are hiring PhD students, postdocs, and interns with a systems background. If you are interested in joining us, get in touch.

Get in Touch