What’s new with FABRIC?

An update from Paul Ruth, FABRIC Principal Investigator

KNIT13 is coming up on September 28 at George Washington University in Washington, DC, co-located with the PAWR PI Meeting. The MERIF Workshop follows at the same venue on September 29 and 30. I hope to see you there.

This year I especially want to hear from you. We're starting to plan the next phase of FABRIC, and KNIT is where we learn what you need that FABRIC doesn't do yet. If there's an experiment you couldn't run, or hardware you wish we had, bring it to DC.

We now list 191 publications that used FABRIC, and I'm sure we're missing some. If one of yours isn't on the list, please let us know.

If you haven't tried LoomAI Hub yet, it's at loomai.net. It's now the recommended way to use FABRIC. You sign in with your FABRIC identity and get drag-and-drop topology building and an AI assistant in the browser, with nothing to install. There's more in the FABRIC News section below.

In this newsletter, you’ll find information about…

Exploring LLM Agents and Decentralized Workflows on FABRIC | Mastering FABRIC: Tips and Tricks Webinar
FABRIC News
Opportunities to collaborate with the FABRIC team

Exploring LLM Agents and Decentralized Workflows on FABRIC | Mastering FABRIC: Tips and Tricks Webinar

September 15, 2026, 3:00-4:30 pm ET

Join us on September 15, 2026, from 3:00–4:30 p.m. ET for the next Mastering FABRIC: Tips and Tricks webinar. In this session, Komal Thareja will share how she is using FABRIC to explore decentralized workflow management and LLM-powered agents through her work on the Swarm project.

Traditional workflow management systems often rely on a centralized authority to assign jobs and manage resources. This webinar will explore an alternative approach in which distributed resources, such as FABRIC sites and virtual machines, act as a swarm of agents that can interact, reach consensus, and allocate jobs among themselves. Komal will discuss the challenges of scaling this approach, the consensus algorithm developed as part of the project, and how LLMs can support agent decision-making when determining where jobs can be most effectively executed.

Attendees will also learn about several approaches for incorporating LLMs into experiments on FABRIC, including hosting a local LLM on a FABRIC virtual machine, using LLMs available through the FABRIC Gateway, and comparing these approaches with a cloud-based LLM service. The session will conclude with a demonstration showing how users can get started with Pydantic AI agents and FABRIC LLM capabilities through a simplified example.

Whether you are interested in AI agents, LLM experimentation, decentralized systems, or exploring new ways to use FABRIC resources, this webinar will provide practical insights and ideas for getting started.

The webinar recording will be shared with all registrants following the event. Those interested can register here.

FABRIC News

Thread the Needle: Advancing AI-Powered Networking and Cybersecurity with FABRIC

Researchers at the University of South Carolina, Dr. Elie Kfoury and Dr. Jorge Crichigno, are using FABRIC to explore how programmable switches, SmartNICs, and other specialized networking accelerators can improve network performance and cybersecurity for next-generation AI applications. Their work uses machine learning on DPUs to detect malicious activity and identify application types within encrypted traffic by analyzing metadata—such as packet timing, sizes, and communication patterns—without decrypting user data. FABRIC is essential to the project because it provides access to advanced networking hardware, large-scale experimental environments, and reproducible Jupyter notebook workflows that allow researchers and students to quickly create, repeat, and share complex experiments. Through their research and educational efforts, the team is preparing the next generation of networking professionals to build faster, smarter, and more secure infrastructure for AI. Their work has received Best Paper Awards at IEEE conferences in 2024 and 2026, with future research focused on running small language models directly on DPUs and improving the efficiency of large-scale AI workloads. See full article linked here: https://www.whatisfabric.net/news-and-blogs/2026-09-03-thread-the-needle-advancing-ai-powered-networking-and-cybersecurity-with-fabric

Educational Use: Using FABRIC for Class Projects

As a national research infrastructure for cutting-edge and exploratory research in networking, distributed computing, and science applications, FABRIC encourages faculty to use the infrastructure for teaching their classes. Training next-generation researchers to imagine and construct new computing and networking experiments at-scale is an important goal for FABRIC. FABRIC has created sample experiments and Jupyter notebooks that are ready to be used by networking and systems classes for instructors.

These experiments include basic ping, network routing, exploring IPv6, TCP analysis, traffic generation, setting up a web server, using Ansible to manage a set of nodes, etc. They are available via GitHub as well as the FABRIC Artifact Manager. Of course, you can create your own assignments for students to do on the FABRIC infrastructure. The FABRIC team provides support from creating an education project, enrolling students, to answering questions from instructors and students. More information can be found here.

FABRIC Office Hours

Connect and troubleshoot with leadership and support

We have made a new office hours system available from the FABRIC portal that allows stakeholders to directly book time with the leadership team and technical team members to discuss anything from the feasibility of their experiments to software questions, to experiment security, to connecting new facilities into FABRIC.

Book an appointment on our scheduling platform.

Solicitations

Funding opportunities that encourage the use of FABRIC

See a list of all solicitations mentioning FABRIC on our website.

Do you have a project idea that would benefit from using FABRIC? The FABRIC team welcomes requests for Letters of Collaboration. To expedite the process, please contact us by filling out the form.

Upcoming Events