Circle
Pluses
Punch card
Arrow

An open AI
community
for all

Hi there 👋 -- we are a community of practitioners working on open-source shareable building blocks (from silicon to open models) empowering humanity to own its own AI. We bring together hackers in the fields of computer architecture, ASIC design, advanced systems, and neural network compilers. Join our thriving community and share, collaborate and innovate with us!

Join the "AI Plumbers" Dev Room

This DevRoom is a celebration of Open Source Low-level AI projects of all types and sizes – the "plumbing" of the AI industry

We aspire for this DevRoom to become as important to the AI community as the Linux Plumbers Conference is to the Linux community.

We are bringing together the top developers working on the essential "plumbing" of the AI industry: hardware accelerators, math kernel libraries, model quantization techniques, low-level inference, fine-tuning engines, distributed and rack-scale computing, and more. Together, we will spend the day discussing core designs and collaborating to solve governance problems.

We are looking for low-level AI core open source project maintainers and committers (such as ggml, llama.cpp and tinygrad, zml, etc.), downstream projects building on top of these (for example, Triton and Dynamo Inference Servers and Docker model runner), as well as end-users of AI stacks to speak about their work and expertise.

If you're hacking on any of these project
and would like to share your experience,
please submit a talk:
Projects
# ggml/llama.cpp
# tinygrad
# zml
# Apache Airflow
# RISC-V based open source AI accelerator hardware
# Inference Servers (Triton, Docker model runner)
# Hugging Face
# Bielik Project
# AIFoundry
KEY DATES AND LOGISTICS
Conference dates: 31st of January & 1st of February 2026
AI Plumbers Devroom: Saturday January 31st, 2026 (whole day)
Submission deadline: December 1st, 2025
Announcement of selected talks: December, 15th 2025
Where: FOSDEM 2025, ULB Campus Solbosch, Brussels, Belgium
Format: Single track, 10 - 20 minute technical session

ABOUT FOSDEM

FOSDEM is a non-commercial event organized by the community for the community. It will take place on the weekend of 31st of January & 1st of February 2026, in Brussels, Belgium. Participation is free and open to everyone. Last year, attendance exceeded 8,000 participants.

FOSDEM is free to attend. There is no registration.

View and join our
open-source
projects

AI Foundry open source projects are Apache
V2 licensed. View the code on our GitHub, and
collaborate with project contributors in our
Discord Server.

ET Platform

Arrow up right
ET is an open-source manycore ASIC platform for parallel computing acceleration.
  • ET Platform is a RISCV, manycore architecture
  • The ET SOC1 contains 1088 compute cores (called minions)
  • Each minion has two rv64imfc RISCV HARTs with vendor-specific vector and tensor extensions
  • For a full understanding of the ET SOC1 architecture check the [ET SOC1 Programmers Reference Manual]
  • Licensed under Apache License, Version 2.0

Inference frameworks

Arrow up right
ET backends for various modern inference frameworks.
  • ggml/llama.cpp
  • tinygrad
  • onnxruntime

Research in Quantization

Arrow up right
A community research project to investigate data-free quantization-aware training.
  • Quantization Aware Training
  • MHAQ
  • tinyMHAQ
Community background

What's happening
in the community

Upcoming podcasts, virtual and physical
community events for AI Foundry.

Past
Upcoming

Al Plumbers Unconference:
San Francisco Edition

Icon calendar

October 25, 2025 Saturday

9:00 AM - 5:00 PM PST

Location icon

SF, USA

Berlin Buzzwords:
AI plumbers conference

Icon calendar

June 15, 2025 Sunday

9:00 AM - 5:00 PM GMT+1

Location icon

Berlin, Germany

FOSDEM fringe:
AI plumbers conference

Icon calendar

Feb 3, 2025 Monday

9:00 AM - 5:00 PM GMT+1

Location icon

Gent, Belgium

Graffiti

Explore AI Plumbers
Insights
Latest Insights background

Stay updated with our community-driven content.
Read about past events and access to summaries
and video replays of prior AI Plumbers Events and
Podcasts.

Vision & Mission Section

Vision
& Mission

Error
Graffiti

Our long-term vision is to help the industry evolve to a structure where model training and deployment for any company are open, transparent, and concise.

We desire to see the industry aligned on the principles of how those systems are structured at
a high level from which every company can
deploy its own engine or system, configured the
way each specific company needs it. This approach will lower risks and improve AI results for all companies and practitioners. Therefore, we foster, build, and support projects that can
become the engineering foundation of such training and deployment systems. We support practitioners willing to benefit from AI in their work and life, specifically models and their fine-tuning and deployment.

We are a community of practitioners building an open-source composable AI ecosystem. By collaborating on standards for everyone, we aim to reduce the complexity of the AI industry.

Our core values

Openness

We don't just mean you can use our tools, we mean you are free to use it however you want.

Integrity

Honesty, ethics, and follow-through foster trust within the community.

Empathy

Our community members are mindful and curious about other people’s motivations and ideas and help each other with compassion and kindness.

Cooperation

Our community seeks win-win solutions with mutually beneficial outcomes.
Background top
Icon star
Icon star
Icon star
Icon star
Icon star
Joining AI Foundry was a crucial decision for Cybergizer. In a post-foundational model era, owning your inference stack isn’t optional!
Avatar
Sergey Sergyenko
CEO at Cybergizer
Logo
Doodle
What people say
Join the AI Revolution background
Join the ai
revolution
today
Vertical text
Computer
Arrows

Faqs

Find answers to your questions about AI Foundry
and our community-driven projects.

Chevron
It may seem like that at first (after all we've got efforts ranging from AI-inference chip design to Quantization Aware Training [QAT] research) but the overarching theme for us is building an inference-focused, economically viable, decentralized, self-governed and developer-centric "Open Source AI" platform. Just like the original Cloud Foundry (from which the inspiration of our name came from) had to tackle a lot of layers at once in order to simplify the cloud application development lifecycle - we have to do the same heavy lift to allow AI practitioners to focus on code rather than infrastructure. At least Cloud Foundry didn't have to design its own alternative to Intel and AMD CPUs!
Chevron
As great Alan Kay once said: "People who are really serious about software should make their own hardware". The problem is that all the successful AI acceleration chips on the market today came from the industry's long focus on training (with inference being more of an afterthought). It is now clear that at least in the enterprise, inference use cases will completely dwarf the training ones. Not only do we need an ML framework that reflects that shift we need it to co-evolve with a chip (same way that CUDA/Pytorch co-evolved with NVidia and tensorflow co-evolved with the TPU). In fact, the combination of the two is so essential and downright critical for the future of business and society that we feel that Open Source is the only way to design the two together.
Chevron
Truth be told, the right way to look at the AI market is before and after the "DeepSeek moment". Not that it wasn't clear to some of us even before, but after the DeepSeek our whole industry got catalyzed around importance of inference and all the efforts (down to the custom chips) that would allow enterprises to have the lowest TCO numbers possible. C[A]IOs started demanding solutions for "OpenAI-like inference APIs around open-weight models at DeepSeek-like TCO". An Open Source platform for something like this has to be built and the time is NOW!
Chevron
AI Plumbers was envisioned to be an interactive conference for all the people building low levels of the AI stack (from hardware to inference engines) - sort of like Linux people have Linux Plumbers Conference. From that standpoint, AI Plumbers is similar to an unconferences since the agenda is fluid and determined by the participants. However, an AI Plumbers involves hands-on work (don't forget to bring your laptop). We partner with other open source and AI conferneces all over the world to make it easier on travel and scheduling for participants, so keep an eye on our events section and don't miss the next one