Google Open Source Blog

This Week in Open Source for October 23, 2026

2026-10-23T11:30:00.000-07:00

by Daryl Ducharme, Open Source Programs Office

A look around the world of open source

What actually sustains an open source community when the cost of generating code drops toward zero, while the human labor required to review, govern, and secure it stays just as demanding? Across this week's reads, researchers, legal scholars, and builders examine the tension between automated volume and human stewardship—from public-infrastructure policy and maintainer bandwidth to provenance transparency, copyright law, and compact models running entirely on local hardware. Because software only serves people when human judgment remains in the loop, protecting maintainer agency and keeping project doors open to real newcomers matter more than ever.

Upcoming Events

🗓️ October 2026

Open Source AI Week 2026 (October 19–24, 2026) — San Francisco Bay Area, California, USA. Now in its final stretch across the Bay Area alongside PyTorch Conference, this decentralized week of community meetups, workshops, and hackathons—with Google DeepMind and Gemma as Platinum Sponsor—wraps up this weekend with tonight's Open Source AI Awards Night (Friday, October 23, 5:00–10:00 PM PT at Fort Mason, San Francisco) and tomorrow's Google Gemma Hardware Hackathon at AGI House (Saturday, October 24, 10:00 AM–5:00 PM PT in Hillsborough, CA).
AGNTCon + MCPCon North America 2026 (October 22–23, 2026) — San Jose, California, USA. Happening this week, this Linux Foundation and Agentic AI Foundation (AAIF) conference focuses on building and scaling production agentic systems with open standards—featuring keynotes by Google's Rao Surapaneni, Sarah Drasner, and Chi Wang, plus sessions from Googlers Alan Blount (on Agent-to-User Interface [A2UI]), Junjie Bu (on interoperable agent discovery), Amit Handa (on agentic commerce), Dominic Farolino (on Model Context Protocol [MCP] Apps + WebMCP), and Prerna Kakkar & Kurtis Van Gent (on EvalBench).
GitHub Universe 2026 (October 28–29, 2026) — San Francisco, California, USA & Virtual. Annual global developer gathering highlighting open source workflows, collaborative security, and multi-model AI-assisted software engineering.

🗓️ November 2026

Open Source in Finance Forum (OSFF) New York (November 4–5, 2026) — New York, New York, USA. Dedicated industry conference examining open source compliance, supply-chain security, and collaborative innovation across regulated financial institutions.
SeaGL (Seattle GNU/Linux Conference) 2026 (November 6–7, 2026) — Seattle, Washington, USA & Virtual. Free, grassroots technical summit dedicated to spreading awareness and knowledge of free and open source software, hardware, and culture.
KubeCon + CloudNativeCon North America 2026 (November 9–12, 2026) — Salt Lake City, Utah, USA. The Cloud Native Computing Foundation's (CNCF) flagship conference uniting adopters and maintainers around Kubernetes and cloud-native infrastructure—featuring over 35 sessions and keynotes from Googlers including Abdel Sghiouar, Jago Macleod, Tim Hockin, Kaslin Fields, John Belamaric, Dawn Chen, Janet Kuo, Benjamin Elder, Antonio Ojea, Lior Lieberman, Michelle Au, and Braydon Kains across Kubernetes Special Interest Groups (SIGs), etcd, containerd, Gateway API, Common Expression Language (CEL) admission policies, and agent sandboxes, plus the Google-sponsored Community Hub.
SFSCON (South Tyrol Free Software Conference) 2026 (November 13–14, 2026) — Bolzano, Italy. One of Europe's longest-running Free Software conferences bringing together public-sector decision-makers and developers to advance digital sovereignty and open infrastructure.

🗓️ December 2026

Yocto Project Virtual Summit 2026.12 (December 1–3, 2026) — Virtual. Community summit for embedded Linux engineers and maintainers building custom, reproducible Linux distributions across edge and Internet of Things (IoT) hardware.
Open Source Summit Japan 2026 (December 7–9, 2026) — Tokyo, Japan. The Linux Foundation's premier Asia-Pacific conference connecting open source developers, kernel maintainers, and enterprise leaders across cloud, embedded Linux, and open AI, including sessions from Google engineers such as Anurag Sinha.
Open Compliance Summit 2026 (December 10–11, 2026) — Tokyo, Japan. Specialized gathering uniting legal, security, and Open Source Programs Office (OSPO) practitioners around OpenChain, Software Package Data Exchange (SPDX), software bills of materials (SBOMs), and global open source supply-chain compliance.

Open Source Reads and Links

[Blog] Open Enables a Lot: The OSI at UNGA81 — At the 81st United Nations General Assembly (UNGA81), the Open Source Initiative (OSI) hosted a panel featuring Google OSPO’s Amanda Casari alongside policy and academic leaders to explore how open source reshapes public infrastructure. Framed by Duane O’Brien’s takeaway that “open by itself does very little, but enables a lot,” what shared scaffolding is needed to turn open licenses into real public value?
[Paper] “AI Slop is DDoSing Open Source”: Understanding the Impact of AI-Generated Contributions on Open Source Sustainability — After hearing questions about AI-generated pull requests online and at conferences for a while now, I came to Google and Oregon State University’s study of 294 repositories looking for what solutions are actually working. Surveying 11 remediation strategies as first-time contributor merge rates drop 18.18%, how can communities filter synthetic noise without closing the door on newcomers?
[Article] Who’s Writing Open Source Code? — Have AI coding agents actually taken over open source repositories, or is human authorship still the norm? Auditing 24,000 commits across 15 critical C infrastructure projects from the first half of 2026, RedMonk’s Stephen O’Grady finds that fewer than 1% carry explicit machine co-author or agent markers—raising a pressing question: how should open source communities build cultural norms around provenance transparency?
[Article] The copyright problem with AI-generated code — In creative circles online, people often argue that prompting AI makes you closer to an art patron or a music producer than the artist—but I had never thought about the copyright realities for software. Since open source licenses rely on human copyright, FreeDOS founder Jim Hall asks a foundational question: can you even license code you didn’t actually author?
[Blog] Launching an opt-in vulnerability-finding service for open-source software — Inspired by Google’s OSS-Fuzz, Anthropic’s new opt-in OSS Scanner offers open source maintainers free, periodic security audits complete with test reproducers and suggested patches running in isolated Docker environments. Giving resource-constrained projects direct access to frontier vulnerability detection is a welcome step forward—how can we keep expanding shared security tooling to help upstream maintainers catch complex bugs early?
[Blog] Do Agent Skills Help? I Ran 72 Trials to Find Out — Evaluating whether AI agent skills beat prompt-and-guess baselines, Karl Weinmeister ran 72 sandboxed trials across 12 scenarios using Harbor and SkillEvaluator—complementing Google’s open source skill-reach suite for catching description collisions across harnesses like AAIF’s Goose and Claude Code. Together, they raise a practical question for maintainers: how do we test both skill discovery and execution quality?
[Blog] Make the PDS Your Own: Customization and Metrics — The Authenticated Transfer Protocol (AT Protocol) is one of my major interests right now, so seeing how people are making its foundations work in practice is fascinating. The latest Reference Personal Data Server (PDS) release adds custom OAuth login branding, 300MB video blobs, and direct-to-Prometheus OpenTelemetry metrics in under 1GB of RAM—what other operational building blocks will help small teams self-host federated infrastructure?
[Article] Google’s new 4B Gemma model cuts speaker-labelling errors on all four benchmarks, two barely — Automated transcripts still struggle when speakers talk over one another in meetings and interviews. Built on Google’s Gemma 4 E4B and open source DiarizationLM framework, the community-trained Apache 2.0 DiarizationLM-Gemma-4-E4B-v1 approaches speaker diarization as a lightweight post-processing pass over existing transcripts—beating an 8B Llama 3 baseline at half the size. Where can compact local models clean up brittle pipeline handoffs in your own workflows?
[Blog] Open-Source Voice AI runs entirely on ESP32-S3 — Some of the most inspiring moments in open source come from individual builders experimenting at the edges of hardware constraints. Anankush Pandit’s GNU General Public License (GPL) Ivy AI packs wake-word detection, speech recognition, a 4-bit language model, and speech synthesis onto an ESP32-S3 microcontroller with just 8MB of pseudo-static RAM (PSRAM)—making us wonder what other everyday devices could run entirely offline?

Which of these stories will you be chatting about at your next meetup or conference? Let us know! Share with us on our @GoogleOSS X account or our @opensource.google Bluesky account.

Introducing the Elephant-Goldfish Model plugin: scaling design-driven AI software development

2026-10-22T11:30:00.000-07:00

by Brian Suchy, Jim Quinn, Elise Ewing, Oliver Newth and Dave Rensin on behalf of the Elephant-Goldfish Model team

Shifting left in the AI coding era

AI-powered velocity is transformative; however, it has silently introduced a potential reliability crisis by also increasing the velocity of “slop,” bugs, and session context collapse. Generative AI tools excel at increasing coding speed, but pure speed without structured guardrails means we are often auto-scaling our mistakes rather than simply writing code faster. We need a paradigm shift, aligning the human need for system comprehension with AI’s capacity for execution.

As co-author Dave Rensin outlines in the article Elephants, Goldfish, and the New Golden Age of Software Engineering, “Design is the new code.”

Understanding the Elephant-Goldfish Model (EGM)

The EGM framework is a metaphor for a powerful way to develop AI-assisted software quickly and safely with humans intentionally in the loop.

The “Elephant” is a context-rich, long-horizon AI session that holds the end-to-end architectural vision and project memory. No code is written here; it is purely dedicated to design, intent, and capturing micro-decisions. The Elephant session is where the AI interviews you and you debate the merits of your choices.

Conversely, “Goldfish” are stateless, zero-context sessions used strictly to validate specifications. Because they share no “blind spots” with the designer, they act as objective critics, probing for gaps and edge cases.

The goal of this method is for the AI to correctly produce the code you wanted the first time after you press Enter.

Announcing the open source Elephant-Goldfish Model plugin

To put this methodology into practice at scale, we are thrilled to announce the Elephant-Goldfish Model plugin. Originally, this plugin was developed internally for Google software engineers, and is now open source. Thousands of Googlers use this plugin every day to ship production Google code. Released under the Apache 2.0 license, the repository provides everything you need to implement this framework. Developers will find customizable Markdown rules, defined subagent personas, and structured review workflows ready to integrate into your own environments.

The plugin’s core goal is to bring the EGM framework directly into Antigravity, with other development environments upcoming. It automates the orchestration of the EGM lifecycle: managing context transfer between Elephants and Goldfish, facilitating validation loops, and rendering design documents, shifting the developer’s role from micromanager to executive reviewer.

The EGM plugin in a nutshell

The EGM lifecycle, automated with the plugin, walks the user through three co-design phases:

Phase 1: Growing the Elephant (`egm-brainstorm`)

Developers load the system context and adhere to a “No Code” rule to co-design features, debate trade-offs, and challenge assumptions.

Phase 2: Drafting with the Elephant (`egm-draft`)

Developers iteratively draft a Markdown design document that covers the problem definition, technical plan, alternatives considered, and a detailed implementation plan that enumerates the specific files to change.

Phase 3: The Goldfish Protocol (`egm-review`)

The Goldfish Protocol shifts the model from co-design to adversarial validation, using stateless AI instances to probe the design for weaknesses. The plugin automates a three-step validation sequence:

Comprehension Check: A stateless Goldfish reads the complete design document and must explain the architecture and intent back to the Elephant. If the Goldfish cannot accurately summarize the plan, it indicates the document is ambiguous and requires further drafting.
Antagonistic Review: Multiple Goldfish instances assume the role of skeptical, antagonistic technical reviewers. They are tasked with identifying potential edge cases, hidden dependencies, and logical gaps. Their critique is funneled back to the Elephant session for resolution.
Implementation Readiness Check: A final Goldfish performs a strict go/no-go check. It verifies that all necessary information—file paths, data structures, and function signatures—is present and unambiguous. This ensures the spec is ready for precise transcription into code.

Human judgment is the artifact

Ultimately, human judgment is the only artifact that truly matters in an AI-saturated world and we want to align the human need for system comprehension with AI’s capacity for execution. We hope this plugin will make it easier for developers of all skill levels to follow a process that more quickly and safely converts the idea they had in their head to a robust and well-built app on their screen.

We invite developers, architects, and AI enthusiasts to contribute to the plugin, customize its validation rules, and help define the future of collaborative software engineering.

To get started, check out the Elephant-Goldfish Model repository on GitHub.

Announcing the verifiable quantum circuit compilation competition on Kaggle

2026-10-21T11:30:00.000-07:00

by Tanuj Khattar and Ryan Babbush, Google Quantum AI

While quantum computing holds the key to solving some of our most complex scientific challenges, current hardware is still limited by qubit counts and gate errors. Quantum error correction can overcome those errors, but it adds substantial overhead to every qubit and gate an algorithm uses. That makes circuit size one of the biggest factors in when a quantum algorithm becomes practical: every qubit and gate we remove lowers the bar that hardware must clear. To bring real-world quantum applications within reach sooner, it is critical to first compress the circuits that power them.

To help bridge this gap, the Google Quantum AI team is launching a competition on Kaggle challenging you to use AI to build smaller, more efficient quantum circuits. The competition is powered by kickmix, our new open source C++20 and Python library in Qualtran for efficiently building and testing large quantum circuits. Our companion paper, Leveraging verifiable structure in quantum circuits to find more efficient compilations with agentic AI, describes the approach in detail.

Turning compilation into an AI agent loop

In the past, optimizing quantum algorithms meant domain experts meticulously tuning circuits by hand and publishing incremental wins one paper at a time. Today, agentic AI can search for more efficient circuits automatically, but such results can only be trusted when candidate circuits can be efficiently verified for correctness. For arbitrary quantum circuits, classical verification is intractable because quantum gates create superpositions over exponentially many classical bitstrings. Fortunately, many of the most expensive subroutines in fault-tolerant algorithms can be built within a structured subclass of circuits, called “kickmix” circuits, which we introduced in our earlier work. Built from reversible classical logic, diagonal phase gates, and measurement-based uncomputation, kickmix circuits map each input bitstring to a single output bitstring and phase without creating new superpositions. A classical computer can therefore run randomized fuzz testing on bitstring inputs in linear time to establish the circuit’s approximate correctness on full quantum states with high confidence.

Randomized fuzz testing does not prove exact equivalence, but most quantum algorithms can tolerate a small, bounded error. In fact, using approximations can cut gate and qubit counts significantly. Finding such approximations by hand is tedious, making it a great task for AI agents. With kickmix in Qualtran, agents can build large circuits quickly and fuzz test them at 60 billion gates per second on a single CPU core.

How the competition works

The competition features three benchmarks, each targeting a major kickmix bottleneck in a useful quantum algorithm in order to call attention to the broad utility of this verification paradigm. For each benchmark, we provide a reference baseline circuit and a standalone verifier, both open source under the Apache 2.0 license. Your goal is to submit a circuit that passes the verifier while minimizing a cost score based on its qubit and gate counts:

Fusion energy: Designing inertial confinement fusion targets requires simulating how quickly charged particles slow down in a hot, dense plasma fuel (Proceedings of the National Academy of Sciences (PNAS)). The main bottleneck is computing the electric potential energy between every pair of particles, which is inversely proportional to their distance. The task is a quantum take on the classic fast inverse square root: given the squared distance x, apply a phase proportional to 1/√x, accurate to a tiny fraction of its exact value.
Optimization: Decoded Quantum Interferometry (Nature) is a quantum algorithm for solving hard combinatorial optimization problems. On one such problem, Optimal Polynomial Intersection (OPI), it finds better solutions than any known efficient classical algorithm. The main arithmetic bottleneck in solving OPI is decoding a Reed-Solomon code in superposition. The task is to build this decoder as a quantum circuit.
Chemistry: Fertilizer production depends on nitrogen fixation, which converts nitrogen gas from the air into ammonia. Industry does this at high temperature and pressure, using about 1–2% of the world’s energy supply, while bacteria do it at ambient conditions with the enzyme nitrogenase, whose active site is an iron-molybdenum cluster called FeMoCo. Understanding how FeMoCo works requires modeling its strongly correlated electrons, a long-standing target for quantum computers. Our recent algorithm (Physical Review X (PRX)) makes that quantum simulation cheaper by classically factorizing the molecule’s Hamiltonian into a compact sum of squares, so the quantum computer needs far fewer steps. At each step, the main bottleneck is a quantum lookup table that loads the parameters of that factorized Hamiltonian in superposition. The task is to co-design both parts: find an accurate factorization that needs fewer steps, and build compact circuits to load its parameters so each step costs less.

Get involved

We have already seen the potential of AI-driven quantum research with Gemini 4 Argon, which is helping researchers at Google accelerate progress across both quantum error correction and algorithm design. We’re excited that the broader research community is embracing this approach too, with simultaneous independent work pairing the kickmix paradigm with AI agents to reduce costs on related subroutines. However, our own internal runs with Gemini models show that all three benchmarks can be further improved beyond the circuits reported to date (and we’ve published a hash of those circuits in our recent manuscript to reveal after the competition), so the competition remains open and interesting.

Could we get more out of a quantum computer simply by asking it to do less? We invite you to help us answer that question.

Visit the competition page on Kaggle to read the full details, explore the resources, and submit your work.

This Week in Open Source for October 15, 2026

2026-10-15T11:30:00.000-07:00

by Daryl Ducharme, Open Source Programs Office

This Week in Open Source for October 15, 2026

A look around the world of open source

After speaking about the Agent2Agent (A2A) protocol and Model Context Protocol (MCP) in Toronto last week (which I recapped in Tag-Team Transmission), open agent boundaries were clearly still on my mind while collecting this week's reads. As AI agents, terminal tools, and embedded micro-apps move deeper into everyday software workflows, a central tension keeps surfacing across the open source community: where do human judgment, interface trust, and maintainer sustainability fit in? Whether teams are wiring up open protocols in the Unix shell, running multimodal retrieval locally on a laptop, or stewarding decades-old kernel infrastructure, the healthiest open source architectures don't replace human agency—they give workers clearer boundaries, inspectable tools, and a genuine seat at the controls.

Upcoming Events

🗓️ October 2026

All Things Open 2026 (October 19–20, 2026) — Raleigh, North Carolina, USA. One of the largest community-focused open source conferences on the U.S. East Coast exploring open source software, AI engineering, and maintainer sustainability—featuring Googlers Andrew Brogdon, Mark Techson, and Marc Dougherty speaking on "Where Is My Futuristic App? Generative UI and the Open Future" and Abdel Sghiouar on "Taming Agentic AI: How to Run Agents-Generated Code Safely on Kubernetes", alongside the Google Community Lounge with the Flutter team.
PyTorch Conference North America 2026 (October 20–21, 2026) — San Jose, California, USA. Two days of open source machine learning innovation across the PyTorch ecosystem—featuring a keynote by Google's Bill Jia ("Workload Fungibility in the Age of Agents") and technical sessions from Googlers Chris Jones, Sandeep Pokkunuri, Rob Mulla, Ankita Luthra, Trinadh Kotturu, Claudio Basile, Qi Zhou, Jago Macleod, and Aleksey Vlasenko on TorchTPU, vLLM, SGLang, Ray, and Kubernetes.
AGNTCon + MCPCon North America 2026 (October 22–23, 2026) — San Jose, California, USA. Linux Foundation and Agentic AI Foundation conference focused on building and scaling production agentic systems with open standards—featuring keynotes by Google's Sarah Drasner and Chi Wang, plus sessions from Googlers Alan Blount (on A2UI), Junjie Bu (on interoperable agent discovery), Amit Handa (on agentic commerce), Dominic Farolino (on MCP Apps + WebMCP), and Prerna Kakkar & Kurtis Van Gent (on EvalBench).
GitHub Universe 2026 (October 28–29, 2026) — San Francisco, California, USA & Virtual. Annual global developer gathering highlighting open source workflows, collaborative security, and multi-model AI-assisted software engineering.

🗓️ November 2026

Open Source in Finance Forum (OSFF) New York (November 4–5, 2026) — New York, New York, USA. Dedicated industry conference examining open source compliance, supply-chain security, and collaborative innovation across regulated financial institutions.
SeaGL (Seattle GNU/Linux Conference) 2026 (November 6–7, 2026) — Seattle, Washington, USA & Virtual. Free, grassroots technical summit dedicated to spreading awareness and knowledge of free and open source software, hardware, and culture.
KubeCon + CloudNativeCon North America 2026 (November 9–12, 2026) — Salt Lake City, Utah, USA. The Cloud Native Computing Foundation's flagship conference uniting adopters and maintainers around Kubernetes and cloud-native infrastructure—featuring over 35 sessions and keynotes from Googlers including Abdel Sghiouar, Jago Macleod, Tim Hockin, Kaslin Fields, John Belamaric, Dawn Chen, Janet Kuo, Benjamin Elder, Antonio Ojea, Lior Lieberman, Michelle Au, and Braydon Kains across Kubernetes SIGs, etcd, containerd, Gateway API, CEL admission policies, and agent sandboxes.
SFSCON (South Tyrol Free Software Conference) 2026 (November 13–14, 2026) — Bolzano, Italy. One of Europe's longest-running Free Software conferences bringing together public-sector decision-makers and developers to advance digital sovereignty and open infrastructure.

🗓️ December 2026

Yocto Project Virtual Summit 2026.12 (December 1–3, 2026) — Virtual. Community summit for embedded Linux engineers and maintainers building custom, reproducible Linux distributions across edge and IoT hardware.
Open Source Summit Japan 2026 (December 7–9, 2026) — Tokyo, Japan. The Linux Foundation's premier Asia-Pacific conference connecting open source developers, kernel maintainers, and enterprise leaders across cloud, embedded Linux, and open AI, including sessions from Google engineers such as Anurag Sinha.
OpenCompliance Summit 2026 (December 10–11, 2026) — Tokyo, Japan. Specialized gathering uniting legal, security, and OSPO practitioners around OpenChain, SPDX, software bills of materials (SBOMs), and global open source supply-chain compliance.

Open Source Reads and Links

[Blog] Meet the A2A CLI: discover, message, and manage agents from your terminal — In Developer Relations, few things beat showing a protocol live in a terminal during a conference talk without a complex web stack in the way. The new Apache 2.0 A2A CLI lets you discover, message, and wrap any stdin/stdout script into a streaming Agent2Agent server with --exec—raising an interesting question: does meeting developers in the Unix shell matter more for open protocol adoption than heavy SDKs?
[Blog] Talking to Agents with a2a (as a human!) — Right after the MCP Dev Summit in Toronto, Marc Dougherty shared a post that uncovered even more about the Agent2Agent (A2A) protocol by flipping the lens from autonomous systems to human-to-agent interfaces with A2UI. Building a2term so agents render interactive terminal confirmation forms before acting raises a compelling question: could open agent protocols also save humans from enterprise UI bloat?
[Article] Dreamforce 2026 — When major enterprise platforms start telling customers they may never need to log into a web UI again because every workflow is now an API, MCP tool, or CLI command, where does the user interface go? RedMonk's Rachel Stephens examines this shift toward headless software and asks a question every open source protocol builder should ponder: when third-party agents mediate our data and workflows, how well does trust actually travel?
[Blog] EmbeddingGemma 2: an open, lightweight multimodal embedding model — Moving RAG setups fully on-device opens up real-world possibilities—from kiosks in low-connectivity areas and offline backup systems to hardware that ships with all of its own help and support docs onboard. With Google DeepMind's Apache 2.0 release of EmbeddingGemma 2 fitting multimodal search into 567MB of RAM, what new resilient tools become possible when devices don't need an internet connection to know how they work?
[Blog] Atproto Web Tile Embeds — Seeing AT Protocol Web Tiles in the wild sparks curiosity about what new publishing patterns become possible when open protocols move beyond text and static images. Nathan Cassell walks through embedding a live, sandboxed 3D Web Tile straight from a personal data server into a blog post—making us wonder what developers might build when interactive micro-apps can travel anywhere without phoning home.
[Report] Navigating Open Source Governance in the Age of Artificial Intelligence: The 2026 State of OSPOs and Open Source Management — Working in an OSPO myself, seeing how other teams operate is essential for improving how we serve as liaisons between our organizations and the broader open source ecosystem. With the Linux Foundation and TODO Group's 2026 survey finding 79% of AI-involved OSPOs shaping model policy and 69% piloting agentic workflows, how can we ensure enterprise AI governance also supports upstream maintainers?
[Report] Celebrating 35 Years of Linux: The Open Source Engine Powering Three Decades of Global Innovation and Infrastructure — It is incredible to see the success of the Linux kernel and distributions across enterprise software development—and to ponder how a project from the early 1990s has kept up with so many technological shifts. As Linux Foundation Research celebrates 35 years of Linux across billions of devices and every top-500 supercomputer, how can the community's model of continuous adaptation guide the next generation of maintainers?

Which of these stories will you be chatting about at your next meetup or conference? Let us know! Share with us on our @GoogleOSS X account or our @opensource.google Bluesky account.

How much does it cost to use open source software?

2026-10-08T11:30:00.000-07:00

by Sophia Vargas, Google Open Source

Open source software is used pervasively, with some estimating that over 90% of codebases contain open source software. Many adopt open source software because it’s free and thus cheaper than writing a proprietary alternative. But as the community saying goes: “open source is free as in puppies, not free as in beer.” Once you or your organization has adopted an open source project, there is still effort (such as engineering time, costs, etc.) associated with ensuring that your version of an open source project is healthy and up to date.

While efforts by the research community have investigated the ongoing costs required to maintain open source software, at Google we needed to better understand the resources required to use thousands of open source software packages.

To better understand this cost, we designed a study to evaluate the effort (in software engineering hours) associated with maintaining open source software packages within Google. We focused on the following questions:

RQ1: How long does it take for an engineer (not a maintainer of the upstream project) to update an open source software package at Google?

RQ2: Can complexity metrics predict the relative effort/amount of time to update a package at Google?

Customization incurs additional cost

The majority of open source package updates (at Google) take four hours or less. However, the long tail could be days to months to fully land a package update. We needed to understand what makes the difference between hours, days, and months.

Our results suggested metrics that could be better predictors of update complexity than others. For example, local patches and customization, the number of package dependencies, and the number of upstream contributors, were better indicators of update complexity than the age of the package, or how many teams were using that package internally.

Study limitations and future work

As this study was completed at Google, we acknowledge there are bias and limitations of our findings, including the following:

Business priority-based sampling: While rigorous, generalizable research studies thoroughly interrogate samples to minimize bias and maximize analytic potential, we designed this study to focus on existing data and minimize new data collection which could impact engineering roadmaps.
We did not dictate tooling: We did not specify what tools (including genAI-enabled features) could or could not be used in the update process, and as such did not investigate the impact of any specific tool.
Bias towards Go: Our collected data had significant bias towards Go. Given the lack of variety in our sample, we could not investigate the effect of language ecosystems, but other internal efforts indicated that language may significantly impact maintenance efforts. We encourage this line of investigation for future studies.
Only active projects: This effort did not include significant data collection on stale or archived packages.

Curious about our approach and findings? Read the full research report: “How much does it cost to maintain a copy of an open source software package (at Google)?”

Adopting Pyrefly as the Python type checker at Google

2026-10-06T11:30:00.000-07:00

by Amer Elsheikh, Python Team

Python is one of the primary languages at Google, powering everything from machine learning research to developer infrastructure and data pipelines. Across hundreds of thousands of files in our monorepo, fast and reliable static type checking is essential for maintaining code safety and developer velocity.

Today, Pyrefly—an open source type checker developed by Meta—is the Python type checker at Google.

Replacing our previous type-checking infrastructure with Pyrefly provided significant speedups for engineers and agents alike, while substantially reducing build-system compute resources.

Moving away from Pytype

For more than a decade, Google relied on Pytype, an internally developed type checker that pioneered Python type checking and collaborated with the open source community to create typeshed.

However, as Python’s typing system evolved rapidly, Pytype faced fundamental challenges. Because it operated by analyzing compiled bytecode rather than source ASTs, keeping pace with modern typing PEPs became an increasing maintenance burden due to bytecode instability across Python releases. In addition, it struggled to deliver the fast turnaround times required for modern development loops.

As detailed in the Pytype update, we decided that Python 3.12 would be the final supported version for Pytype, leading us to evaluate and adopt modern open source alternatives.

Why Pyrefly?

Pyrefly incorporates years of collective lessons from earlier tools across the Python typing ecosystem. When evaluating candidates to succeed Pytype, Pyrefly quickly stood out across three key areas:

Performance

Written in Rust, Pyrefly is designed for high throughput and lazy, parallel evaluation. In our internal benchmarks across various Google projects, it proved to be an order of magnitude faster than Pytype, while scaling smoothly across large dependency graphs.

Typing spec conformance

Pyrefly achieves strong conformance—scoring roughly 97% on the official typing conformance test suite—and is actively maintained to track new Python versions and typing PEPs. This comprehensive language support makes Python version upgrades across our monorepo smoother.

In addition to supporting new syntax, Pyrefly provides strict type safety in areas where Pytype was historically permissive. A notable example is unsound unions: Pytype allowed passing a value of a union type (such as int | None or int | str) to a function expecting a specific type (such as int), as long as at least one type in the union was compatible. However, Pyrefly enforces sound union checking, catching these mismatches:

# foo.py
def process_id(x: int) -> None:
    ...
def get_id() -> int | None:
    ...
val = get_id()
# Accepted by Pytype, but rejected by Pyrefly:
process_id(val)

Actionable error diagnostics

Pyrefly provides Rust-style compiler diagnostics, displaying the offending code snippet with inline annotations that point directly to the root cause of type mismatches, sometimes with suggested fixes. E.g., for the foo.py example above, Pyrefly produces:

ERROR Argument `int | None` is not assignable to parameter `x` with type `int` in function `process_id` [bad-argument-type]
 --> foo.py:8:12
  |
8 | process_id(val)
  |            ^^^
  |
  The declared type does not allow `None`. Consider narrowing the value with an `is not None` check.

Performance and infrastructure impact

Switching to Pyrefly brought measurable improvements across Google's developer ecosystem and infrastructure:

Up to 98% faster incremental rebuilds: In developer edit-and-rebuild workflows (with a warm daemon during active editing cycles), Pyrefly delivered up to a 98% latency reduction across various targets. On large machine learning targets, type-checking times dropped from minutes to seconds.
>90% critical-path reduction in clean builds: In cold-cache benchmark suites across major libraries and models, Pyrefly consistently reduced the type-checking share of the build critical path by 90% to 99%, eliminating a long-standing bottleneck in our build pipelines.
>80% compute hardware savings: Pyrefly reduced Google’s daily peak compute occupancy for Python type checking by more than 80%, saving thousands of machine cores every day.

To illustrate this impact on developer workflows, the chart below compares total edit-and-rebuild latency between Pytype and Pyrefly across targets of varying sizes.

Developer feedback

Beyond aggregate metrics, Pyrefly improved the day-to-day development loop, allowing engineers and AI coding agents to catch bugs faster.

Here is what engineers across Google have shared about their experience:

"I haven't waited for a Pytype action to complete since our project switched [to Pyrefly]. It is doubly awesome for agentic coding. Type checking is faster than running the tests now." — Peter Hawkins, JAX Tech Lead

"I really like that Pyrefly gives super clear errors that point out exactly what's wrong." — Yotam Doron, Gemini Large Scale Pretraining

“Investing in Python tooling pays huge dividends for research velocity: Pyrefly keeps our experimental iterations fast and catches subtle bugs early with clear, actionable errors.” — Tom Ward, GDM Science

Looking forward

Adopting Pyrefly highlights the value of uniting behind shared open source developer tooling. We are deeply grateful to the Pyrefly team for their rapid turnaround and responsiveness on upstream issues throughout our rollout. We look forward to continuing our collaboration and contributing to the Python open source community.

To learn more about Pyrefly or try it in your own projects, visit the Pyrefly website and the Pyrefly GitHub repository.

This Week in Open Source for October 2, 2026

2026-10-02T11:30:00.000-07:00

by Daryl Ducharme, Open Source Programs Office

This Week in Open Source for October 2, 2026

A look around the world of open source

As AI becomes part of everyday software development, the most compelling questions in open source are shifting from the models themselves to everything surrounding them. I’m always drawn to how technology affects individuals, and especially the workers who use and maintain it, for better or for worse (I definitely prefer better). That’s why I’m convinced the next phase of open-source AI won’t be won by chasing bigger benchmarks, but by building the open infrastructure, transparent local tooling, and community norms that keep developers and maintainers in control. This week’s reads explore what that looks like in practice.

Upcoming Events

🗓️ October 2026

ValkeyConf 2026 (October 5, 2026) — Prague, Czechia. Dedicated community conference advancing the open source Valkey in-memory data store, featuring a keynote by Valkey TSC member and Google Cloud engineer Jacob Murphy on open source in-memory data structures and scaling high-performance workloads.
MCP Dev Summit Toronto (October 5–6, 2026) — Toronto, Ontario, Canada. Dedicated Linux Foundation developer summit advancing the Model Context Protocol (MCP) and open agentic interoperability standards. Join Google OSPO's Daryl Ducharme (that's me!) on Tuesday, October 6 (11:00 AM EDT) for "Tag-Team Transmission: Navigating A2A and MCP for Optimum Orchestration," examining how the Agent2Agent (A2A) protocol and MCP interoperate across multi-agent architectures.
Open Source Summit Europe (October 7–9, 2026) — Prague, Czechia. The premier Linux Foundation gathering in Europe celebrating the 35th anniversary of Linux and connecting developers, technologists, and community leaders—including a keynote by Google OSPO's Erin McKean on Docsy and technical documentation for humans and AI agents.
Community Over Code (October 11–14, 2026) — Glasgow, Scotland. The flagship Apache Software Foundation conference bringing together project maintainers, committers, and users to collaborate on open governance, data architecture, and community-led software development.
BazelCon 2026 (October 13–15, 2026) — Amsterdam, Netherlands. Annual gathering of the Bazel community uniting build system engineers, maintainers, and ecosystem contributors around fast, reproducible, multi-language software builds.
All Things Open 2026 (October 19–20, 2026) — Raleigh, North Carolina, USA. One of the largest community-focused open source conferences on the U.S. East Coast exploring open source software, AI engineering, and maintainer sustainability—featuring a Google keynote on Generative UI and the Open Future alongside the Google Community Lounge with the Flutter team.
PyTorch Conference North America 2026 (October 20–21, 2026) — San Jose, California, USA. Two days of open source machine learning innovation covering training, inference, compiler toolchains, and hardware heterogeneity across the PyTorch ecosystem.
AGNTCon + MCPCon North America 2026 (October 22–23, 2026) — San Jose, California, USA. Linux Foundation conference focused on building and scaling production agentic AI systems with open standards, observability, and control, featuring a keynote by Google's Rao Surapaneni.
GitHub Universe 2026 (October 28–29, 2026) — San Francisco, California, USA & Virtual. Annual global developer gathering highlighting open source workflows, collaborative security, and AI-assisted software engineering.

🗓️ November 2026

Open Source in Finance Forum (OSFF) New York (November 4–5, 2026) — New York, New York, USA. Dedicated industry conference examining open source compliance, supply-chain security, and collaborative innovation across regulated financial institutions.
KubeCon + CloudNativeCon North America 2026 (November 9–12, 2026) — Salt Lake City, Utah, USA. The Cloud Native Computing Foundation's flagship conference uniting adopters and maintainers around Kubernetes, platform engineering, and cloud-native infrastructure.
SFSCON (South Tyrol Free Software Conference) 2026 (November 13–14, 2026) — Bolzano, Italy. One of Europe's longest-running Free Software conferences bringing together public-sector decision-makers and developers to advance digital sovereignty and open infrastructure.

Open Source Reads and Links

[Article] Mila and Mozilla announce new initiative to build trustworthy open source AI for everyone, with Canadian government support — As a Canadienthusiast, seeing Mozilla and Mila partner with the Canadian government on an open source AI foundation layer immediately grabbed my attention. The crossover of open source and public-sector sovereign systems seems clear on the surface, yet real-world adoption reveals unexpected decisions. How do we make open source and open models useful to governments—is it in the open source infrastructure around them?
[Article] Google’s open source EnvHarness lets AI agents train against environments that evolve with them — As AI adoption matures, the need for solid infrastructure around AI is becoming more evident. It is nice to see Google Research’s work leading to open source tools like EnvHarness, which adapts training sandboxes to an agent’s failure modes—and raises the question of how our evaluation infrastructure must evolve alongside the agents themselves.
[Blog] Inside LLM Inference: Every Calculation from Text to Token using Gemma 4 12B — LLM inference often feels like an opaque black box, making Amulya Bhatia’s calculation-by-calculation trace of a token moving through Google’s open source Gemma 4 12B fascinating to ponder. Seeing how hybrid attention bounds memory under the hood prompts a broader question: how differently do we design AI systems when we actually understand the math happening inside the model?
[Video] 100% Local RAG Without Internet: Qdrant Edge and Google LiteRT — GDE Tarun R Jain pairs Google LiteRT and Gemma 4 with Qdrant Edge to run retrieval-augmented generation 100% offline. It prompts an interesting question for system design: how differently do teams experiment with their knowledge bases when local prototyping is free—and how much of what we default to the cloud could stay on the edge?
[Post] UISurf: An Operator-Centric Multi-Agent Platform for Observable and Cross-Environment UI Automation — UISurf explores orchestrating UI agents across Web, Desktop, and Mobile boundaries using the open Agent2Agent (A2A) protocol. Beyond the tool itself, its lessons on sandboxing and human-in-the-loop oversight are worth pondering: when autonomous agents coordinate across environments, what level of observability do human operators actually need to stay in control?
[Article] PS5 Linux lead quits as open-source projects have become “a bunch of noobs using LLMs” that “they don’t even understand” — When maintainers leave high-profile projects like PS5 Linux, it is important to look at the reasons. In the age of AI-assisted software development, unvetted LLM code and shifting bounty incentives are straining both maintainers and community trust. What can we learn from these events as we update open source and AI governance best practices to keep projects safe, secure, and viable?

Which of these stories will you be chatting about at your next meetup or conference? Let us know! Share with us on our @GoogleOSS X account or our @opensource.google Bluesky account.

Standardizing fine-grained access control in Apache Iceberg's REST catalog

2026-10-01T11:30:00.000-07:00

by Talat Uyarer & Sung Yun, Google Cloud

The Apache Iceberg community recently merged a change to the REST Catalog specification that gives lakehouses something they have long been missing: a standard, engine-neutral way to express column masking and row filtering. It is a small change, but it addresses a structural gap in how open table formats handle data governance. This post walks through the problem, the design, and why the details matter.

The problem: there is no server in the read path

In a traditional database, access control is straightforward because there is exactly one door. Every query passes through the database server, the server knows who is asking, and if policy says "this user only sees the last four digits of the card number," the server applies that policy before returning results. One process, one enforcement point.

Apache Iceberg deliberately removed that door. The data is Parquet files in object storage, and any engine—such as BigQuery, Spark, Trino, Flink, PyIceberg, or DuckDB—can read those files directly. This design is what makes Iceberg fast and interoperable. But it also means that once an engine asks the catalog "where is the payments table?" and receives the metadata pointer, it has full access to the files. The catalog can grant or deny access to the whole table, but it has had no vocabulary for "access granted, but mask the email column" or "access granted, but only on rows where region = 'US'."

Until now, Iceberg had no built-in answer to this. The REST spec did offer coarser instruments like credential vending, which controls storage access per table, and server-side scan planning, which lets a catalog withhold entire data files. But neither can express "mask this column" or filter rows that are not already physically separated into their own files. So users who needed fine-grained access control had to step outside the open protocol entirely. They could adopt a vendor-provided client that understands its proprietary policy format, or route every read through a vendor's proxy service, giving up the direct-to-storage performance that motivated the use of Iceberg in the first place. It also quietly undermines Iceberg's core promise: the moment governance requires a specific vendor's client, the table is no longer open to any engine.

The new read-restrictions field in the REST Catalog spec is the community standardizing that vocabulary.

The mechanism

When an engine calls loadTable, the response may now include an optional read-restrictions object:

Read restrictions are expressed in two fields:

required-row-filter: a standard Iceberg predicate expression. Rows for which it evaluates to false must not appear in the result, and no information derived from them may be included.
required-column-projections: a list of columns, identified by field ID, each with a transformation the reader must apply before returning values.

One evaluation rule ties them together: the row filter is evaluated against the original, untransformed column values, and projections are applied to the rows that survive. This ordering is what makes the two features composable. A policy can filter on region = 'US' and also mask region in the output, and the filter still works. If masking ran first, any policy that filtered on a masked column would silently break.

Note what the catalog is doing here. It evaluates the access policy server-side—it knows the caller's identity from the authentication token—and returns only the result of that evaluation. The policy itself, with its roles, tags, and governance model, never crosses the wire. The engine does not need to understand how any particular catalog models governance. It needs to understand nine actions and a predicate.

The whole architecture fits in one sequence:

The catalog stays the policy decision point; the trusted engine becomes the policy enforcement point; the end user never holds storage credentials. Sections below unpack the three load-bearing details in this picture: the enforcement ordering, the fail-closed branch, and the trust boundary.

The nine masking actions

The specification defines a closed set of nine masking actions, each with an exact, per-type definition. The goal is cross-engine consistency: Spark, Trino, and PyIceberg must produce identical output for any given masking action. These actions are being implemented in iceberg-core, providing engines with spec-compliant transformations out of the box rather than requiring them to reimplement byte-level logic independently.

Below, the actions are grouped by the analytical utility of the resulting masked data:

Preserve the shape, hide the value

mask-alphanum—digits become n, other characters become x, with a small allowlist of punctuation (( ) , . - @) preserved. iceberg16112018@apache.org becomes xxxxxxxnnnnnnnn@xxxxxx.xxx—recognizably an email address, but not whose.
show-first-4 / show-last-4—preserve four code points, and apply mask-alphanum to the rest. 4111-1111-1111-4444 becomes nnnn-nnnn-nnnn-4444, the familiar customer-support view of a card number.

Hide everything

replace-with-null—the value becomes NULL. Only valid for optional fields; a server must not return it for a required field, and a reader that receives one must fail the query.
mask-to-fixed-value—the value becomes a type-specific constant (0, "XXXXXXXX", the epoch, an all-zero UUID, an empty list, and so on, each spelled out in the spec). Uniquely among the actions, this one also replaces NULL inputs—even the null-or-not bit is hidden.

Reduce precision

truncate-to-year / truncate-to-month—2024-07-15 becomes 2024-01-01 or 2024-07-01. Cohort analytics keeps working; identifying individuals gets harder.

Allow joins, hide values

sha-256-global—deterministic SHA-256, with exact byte-encoding rules per input type. The same input always produces the same output, everywhere, so GROUP BY user_id and joins across tables on a hashed key still work. The cost of that determinism is that hashed values can be tested against precomputed guesses—this is pseudonymization, not encryption.
sha-256-query-local—the same hash, salted with a fresh, cryptographically random salt (at least 16 bytes) per query. Values remain consistent within a single query, so self-joins and aggregation work, but cannot be correlated across queries, and precomputed-guess attacks no longer apply.

This last pair is a nice piece of design: the tradeoff between linkability and privacy, expressed as two enum values that a policy author chooses between per column.

Every action produces a value of the same type as its input—masked strings are strings, truncated dates are dates—so restrictions never change the schema an engine plans against. Queries do not need rewriting; values simply arrive transformed.

Fail-closed by design

The most consequential sentence in the specification is this one:

If a trusted reader that supports read-restrictions cannot apply any returned restriction, it must fail the query and must not silently return raw, partial, or empty results.

Consider the failure modes. A catalog sends an action added in a future spec version that the engine does not recognize. Or an expression type it cannot evaluate. The convenient behavior would be to skip what it does not understand and return the data. The spec rules this out: unrecognized action, fail; unparseable filter, fail; duplicate field ID in the projections, fail. Every ambiguity resolves to "no data" rather than "raw data." This is the right default for an access control mechanism, and it is also the one implementers would be tempted to soften—which is exactly why it is normative in the spec rather than left to judgment. It is also what allows the action vocabulary to grow in future versions without older engines becoming silent leak vectors.

A few prohibitions in the spec reward a closer look, because each one closes a subtle correctness hole:

No projections on map keys. Masking a map's keys can collapse two keys into one, or produce null keys, which engines silently coalesce or reject—data corruption presented as privacy. The spec bans it outright.
No projection on both a nested type and a field inside it. Masking a struct and also a field within that struct has no well-defined order of operations, so the spec refuses to define one: servers must not send it, and readers must reject it.
Everything references field IDs, never column names. This is standard Iceberg discipline: if ssn is renamed to national_id, the policy remains bound to the same physical column. A name-based policy would silently detach on rename—the worst possible failure mode for access control.

The trust model, stated plainly

All of this is enforced by the reader. The catalog hands the engine the file locations along with the restrictions, and a client that chooses to ignore the restrictions can read the raw files. So what is this actually protecting?

The specification is explicit: this mechanism assumes a trust relationship between the catalog and the engine, and how that trust is established is deliberately out of scope. The intended deployment is one where a platform team's engines—the shared Spark and Trino clusters—are trusted enforcement points that hold storage access (for example, through credential vending), while end users only ever talk to those engines and never hold storage credentials themselves. The trusted engine becomes the enforcement point, playing the role the database server played in the traditional architecture. The difference is that its behavior is now defined by a common, open protocol rather than by N proprietary integrations.

In other words, read restrictions do not protect data from the engine; they let the catalog direct a trusted engine to protect data from the engine's users. For genuinely untrusted readers, the coarse-grained model still applies: they are restricted to vended credentials where the storage access granted to the user aligns with the data permission of the user.

One operational subtlety deserves attention: restrictions are per-response and per-identity. The same loadTable call made by a different principal—or by the same principal later—may return different restrictions, and the restrictions attach to every read performed with that response, including subsequent planTableScan and fetchScanTasks calls. The spec therefore requires that the response not be cached outside its authentication scope. If your platform caches loadTableResponse, that cache is now security-sensitive and worth an audit.

What is still missing

Read restrictions are a foundation, not the finished building. It is worth being honest about the gaps between this specification and complete fine-grained access control:

The action vocabulary is fixed, because Iceberg Expressions are not implemented yet. Nine actions cover the common masking policies, but they are a deliberately closed set: a policy like "apply this custom redaction function" cannot be expressed, and the row filter is limited to predicates—comparisons that produce true or false. The path to generalizing this already exists on paper: the Iceberg Expressions specification, proposed by Ryan Blue and adopted in mid-2026, defines a portable structure for value expressions—constants, field references, and calls to well-defined functions or SQL UDFs. Once engines can evaluate those expressions, a catalog could return arbitrary transformations instead of choosing from an enum. Today no engine implements general expression evaluation, which is why the initial design confines itself to a small vocabulary that can be specified bit-for-bit.
There is no policy definition—deliberately. The specification standardizes the result of policy evaluation, never the policy itself. How an organization expresses "analysts see masked PII, auditors see everything"—the roles, tags, rules, and administrative APIs—remains entirely the catalog vendor's domain, and the assumption is that it stays there. Only the consequences of a policy are portable across engines; the policy definition is not. Whether communities eventually want a portable policy format is an open question the spec does not attempt to answer.
Trusted clients are asserted, not proven. As the trust-model section noted, how a catalog establishes that a caller is a trusted, enforcing engine is out of scope. In practice that trust is deployment configuration—service identities, network boundaries, and which principals receive vended credentials. There is no attestation mechanism in the protocol by which an engine proves it enforces restrictions; the trusted-client mechanism is an assumption the platform operator must make true.

None of these gaps undermines the design—each is a deliberate scoping decision that kept the proposal small enough to reach consensus—but they define the roadmap for what "complete" fine-grained access control in the open lakehouse still requires.

Why this matters

The specification change itself does not ship enforcement; that work in the engines begins now, starting with the default actions. But the shape of the design is right in three ways:

It picks the honest enforcement point. In an architecture with no server in the read path, the trusted engine is the only place enforcement can live without giving up direct storage reads. The spec accepts that constraint explicitly rather than obscuring it.
It standardizes the narrow waist. Catalogs keep their own rich policy engines—roles, tags, and attribute-based rules. Engines implement nine actions and a predicate evaluator, once. N×M becomes N+M.
It is fail-closed everywhere, which is what makes the vocabulary safely extensible.

The pattern—the catalog evaluates policy against the caller's identity and returns a small, closed vocabulary of obligations that the client must enforce or fail—is a useful template, and it would not be surprising to see more of the governance surface expressed this way over time.

The proposal was approved on the Apache Iceberg dev list with eight binding +1 votes and no objections. This is a strong signal of consensus across the many companies and open source communities that participate in the project. Consistent, engine-independent enforcement of fine-grained policies is a property the ecosystem has long wanted; it now has a specification for it, and the interesting work of implementing it in engines and catalogs is underway. If you work on either, the dev list is the place to get involved.

The full schema is available in rest-catalog-open-api.yaml under ReadRestrictions. View the full vote thread.

New public datasets available in Google Cloud Lakehouse

2026-09-30T11:30:00.000-07:00

by Daniel Rodrigues & Alex Stephen, Google Cloud

Did you know the Google Chrome Wikipedia page received over 36 million page views in 2025, or that Chromium logged over 3,700 commits in April 2010? If you have needed large, real-world datasets to benchmark query engines on Apache Iceberg, Google Cloud’s Lakehouse team is excited to announce the release of new public datasets in Google Cloud Lakehouse to help you explore and analyze open data at scale.

Google Cloud’s Lakehouse provides a high-performance storage catalog using Apache Iceberg as its open table format. By decoupling storage from compute, it enables you to use your preferred query engines—such as BigQuery, Apache Spark, or Trino—while managing data in an open format to avoid vendor lock-in.

These new public datasets are designed to help you explore the Apache Iceberg ecosystem and begin working immediately with real-world data. We are providing access to some of the most popular BigQuery public datasets, including Wikipedia pageviews and GitHub commit histories. Let’s dive into how you can start querying them.

Prerequisites

A Google Cloud project (for authentication).
Standard Google Application Default Credentials (ADC) configured in your environment.

Explore metadata with PyIceberg

To explore table metadata using PyIceberg, install the required Python packages in a virtual environment:

sudo apt-get install python3-venv
python3 -m venv public_data_lakehouse
source public_data_lakehouse/bin/activate
pip install google-auth
pip install pyiceberg
pip install pyarrow

After installing the required packages, you can inspect a table’s schema using the Python script below. Here is how to inspect the table containing Wikipedia page view data from 2016:

from pyiceberg import catalog as pyiceberg_catalog

PROJECT = "<YOUR_PROJECT_ID>"
NAMESPACE = "wikipedia"
TABLE = "pageviews_2016"

def load_catalog() -> pyiceberg_catalog.Catalog:
  properties = {
      "type": "rest",
      "uri": "https://biglake.googleapis.com/iceberg/v1/restcatalog",
      "warehouse": "gs://lakehouse-public-data",
      "auth": {"type": "google"},
      "header.X-Iceberg-Access-Delegation": "vended-credentials",
      "header.x-goog-user-project": PROJECT,
  }
  return pyiceberg_catalog.load_catalog(PROJECT, **properties)

def main() -> None:
  catalog = load_catalog()
  table = catalog.load_table((NAMESPACE, TABLE))
  print(f"Schema for `{NAMESPACE}.{TABLE}`:", table.schema())

if __name__ == "__main__":
  main()

Note: Replace <YOUR_PROJECT_ID> with your actual Google Cloud Project ID. This is required for the REST catalog to authenticate your quota usage, even for free public access.

PySpark and Managed Service for Apache Spark

Now that we have explored the table’s schema, we can use a managed, serverless Spark notebook to query the tables. This provides the speed and flexibility of Apache Spark without creating or managing a cluster. You can run this script in Google Cloud's serverless Managed Service for Apache Spark (formerly Dataproc):

from google.cloud.dataproc_v1 import Session
from google.cloud.dataproc_spark_connect import DataprocSparkSession

PROJECT_ID = "<YOUR_PROJECT_ID>"
spark_catalog = "lakehouse-public-data"

session = Session()
session.runtime_config.properties = {
  "spark.sql.extensions": "org.apache.iceberg.spark.extensions.IcebergSparkSessionExtensions",
  f"spark.sql.catalog.{spark_catalog}": "org.apache.iceberg.spark.SparkCatalog",
  f"spark.sql.catalog.{spark_catalog}.type": "rest",
  f"spark.sql.catalog.{spark_catalog}.uri": "https://biglake.googleapis.com/iceberg/v1/restcatalog",
  f"spark.sql.catalog.{spark_catalog}.rest.auth.type": "org.apache.iceberg.gcp.auth.GoogleAuthManager",
  f"spark.sql.catalog.{spark_catalog}.io-impl": "org.apache.iceberg.gcp.gcs.GCSFileIO",
  f"spark.sql.catalog.{spark_catalog}.header.X-Iceberg-Access-Delegation": "vended-credentials",
  f"spark.sql.catalog.{spark_catalog}.warehouse": "gs://lakehouse-public-data",
  f"spark.sql.catalog.{spark_catalog}.header.x-goog-user-project": PROJECT_ID,
}

spark = (
   DataprocSparkSession.builder
     .appName("Lakehouse Public Data Demo")
     .dataprocSessionConfig(session)
     .getOrCreate()
)
spark.conf.set("spark.sql.defaultCatalog", spark_catalog)

With the session active, you can query monthly Wikipedia view counts for BigQuery-related articles by executing the following code in a new cell:

df1 = spark.sql("""
SELECT
  title,
  wiki,
  SUM(views) AS total_views
FROM wikipedia.pageviews_2026
WHERE datehour >= TIMESTAMP '2026-01-01 00:00:00'
  AND datehour <  TIMESTAMP '2026-02-01 00:00:00'
  AND LOWER(title) LIKE '%bigquery%'
GROUP BY title, wiki
ORDER BY total_views DESC
LIMIT 20
""")
df1.show(10)

Or, you can retrieve recent GitHub commits referencing Iceberg using the following snippet:

df2 = spark.sql("""
SELECT
  commits.commit,
  commits.subject,
  commits.message,
  commits.author.name AS author_name,
  timestamp_seconds(commits.committer.date.seconds) AS commit_time,
  repo_name
FROM github_repos.commits AS commits
WHERE LOWER(commits.subject) LIKE '%iceberg%'
   OR LOWER(commits.message) LIKE '%iceberg%'
ORDER BY commit_time DESC
LIMIT 50
""")
df2.show(10)

Start building today

These datasets were imported from their BigQuery counterparts and transformed using Apache Iceberg as the table format and Parquet as the data file format. Our goal is to lower the entry barrier so you can learn and explore Apache Iceberg with your favorite query engine without managing any infrastructure. To get started with building an open, managed, and high-performance Iceberg lakehouse, visit the Google Cloud Lakehouse page.

This Week in Open Source for September 24, 2026

2026-09-24T11:30:00.000-07:00

by Daryl Ducharme, Open Source Programs Office

This Week in Open Source for September 24, 2026

A look around the world of open source

Open source underpins an $8.8 trillion global economy, yet sustaining it requires moving beyond voluntary charity and philosophical appeals. This week’s reads examine how the ecosystem is adapting to modern economic and AI pressures—from proposing package-registry royalties and quantifying 2–5x net value for regulated utility grids, to turning autonomous AI security incidents into hardened supply-chain defenses that empower open-science breakthroughs like NASA and IBM’s Lunar Foundation Model.

Upcoming Events

🗓️ October 2026

MCP Dev Summit Toronto (October 5–6, 2026) — Toronto, Ontario, Canada. Dedicated Linux Foundation developer summit advancing the Model Context Protocol (MCP) and open agentic interoperability standards. Join Google OSPO's Daryl Ducharme on October 6 for "Tag-Team Transmission: Navigating A2A and MCP for Optimum Orchestration," examining how the Agent2Agent (A2A) protocol and MCP interoperate across multi-agent architectures.
Open Source Summit Europe (October 7–9, 2026) — Prague, Czechia. The premier Linux Foundation gathering in Europe celebrating the 35th anniversary of Linux and connecting developers, technologists, and community leaders across open AI, embedded systems, and digital trust.
Community Over Code (October 11–14, 2026) — Glasgow, Scotland. The flagship Apache Software Foundation conference bringing together project maintainers, committers, and users to collaborate on open governance, data architecture, and community-led software development.
All Things Open 2026 (October 19–20, 2026) — Raleigh, North Carolina, USA. One of the largest community-focused open source conferences on the U.S. East Coast exploring open source software, AI engineering, DevSecOps, and maintainer sustainability.
GitHub Universe 2026 (October 28–29, 2026) — San Francisco, California, USA & Virtual. Annual global developer gathering highlighting open source workflows, collaborative security, and AI-assisted software engineering.

🗓️ November 2026

Open Source in Finance Forum (OSFF) New York (November 4–5, 2026) — New York, New York, USA. Dedicated industry conference examining open source compliance, supply-chain security, and collaborative innovation across regulated financial institutions.
KubeCon + CloudNativeCon North America 2026 (November 9–12, 2026) — Salt Lake City, Utah, USA. The Cloud Native Computing Foundation's flagship conference uniting adopters and maintainers around Kubernetes, platform engineering, and modern cloud infrastructure.
SFSCON (South Tyrol Free Software Conference) 2026 (November 13–14, 2026) — Bolzano, Italy. One of Europe's longest-running Free Software conferences bringing together public-sector decision-makers and developers to advance digital sovereignty and open infrastructure.

Open Source Reads and Links

[Blog] Nobody pays for open source. We can force them to. — Companies already spend over $1 billion a year on open source, but that money goes to supply-chain mirror and security vendors rather than the maintainers writing the code. Laurie Voss breaks down why voluntary charity and license changes always fail, arguing that commercial package registries and mirror vendors should pay automatic royalties down the dependency tree to long-tail maintainers.
[Article] AI Is reshaping open source software and straining the systems that sustain it — As conversations grow around moderating the speed of AI development across the industry, it is critical to watch the friction points where AI and open source overlap—particularly as an influx of AI-generated code and vulnerability discovery strains maintainers of an $8.8 trillion ecosystem. Funding remains front and center, requiring not just raw capital but continuous monitoring for effectiveness in sustaining the human governance, consensus-building, and non-coding infrastructure that AI cannot replace.
[Report] LF Energy Research Finds Open Source Software Can Deliver 2-5x Greater Net Value for Grid Operators — While open source software and public utilities seem like natural philosophical allies as public goods, philosophical appeals routinely fail in regulated infrastructure sectors where operators are bound by strict reliability mandates and ratepayer accountability. This new LF Energy report demonstrates how to bridge that gap: by translating collaborative "make together" governance into an auditable benefit-cost framework showing two to five times greater net value over proprietary procurement.
[Article] AI Is at a Turning Point — High-profile incidents of autonomous AI agents breaching Hugging Face or social-engineering open source maintainers inevitably dominate headlines and amplify purely negative narratives around AI. However, treating these security failures as concrete post-mortems—exposing how standard reinforcement learning incentivizes deceptive subgoals—is exactly what allows open source communities to establish hardened supply-chain defenses and safely advance high-impact, positive AI use cases.
[Article] IBM and NASA Release Open-Source AI Model to Support Lunar Exploration — Seeing open source, AI, open data, and space science converge is genuinely inspiring, especially when it demonstrates that open scientific models are only as useful as the open datasets beneath them. While NASA and JAXA planetary archives have long been public, IBM and NASA's release of the Lunar Foundation Model alongside the first harmonized, machine-learning-ready lunar dataset (unifying 30+ multi-instrument layers) proves that transforming raw open data into shared, interoperable infrastructure is what unlocks domain discovery.

Which of these stories will you be chatting about at your next meetup or conference? Let us know! Share with us on our @GoogleOSS X account or our @opensource.google Bluesky account.

Google Cloud: Investing in the future of PostgreSQL — 2026 highlights

2026-09-23T11:30:00.000-07:00

by Dilip Kumar, Cloud SQL for PostgreSQL

Google Cloud is committed to open source, and PostgreSQL is a cornerstone of managed database offerings, including Cloud SQL and AlloyDB.

Continuing our work with the PostgreSQL communities, we've been contributing to the core engine and participating in the patch review process. Below is a summary of that technical activity between January 2026 and September 2026, highlighting our efforts to enhance the performance, stability, and resilience of the upstream project and ecosystem. By strengthening these core capabilities, we aim to drive innovation that benefits the entire global PostgreSQL ecosystem and its diverse user base.

Our technical contributions in this period have focused on enhancing core engine performance, introducing features for logical replication, fixing critical bugs, and improving upgrade resilience. We also continue to invest in the PostgreSQL ecosystem by addressing bugs in widely used extensions.

Technical contributions: January 2026 – September 2026

Our contributions this cycle span four key areas:

Logical replication and conflict management: Paving the way for active-active replication, enhanced conflict logging, and catalog safety.
Core engine performance and reliability: Eliminating tuple-level overhead in sequential scans and making promotion timeouts accurate.
Core catalog, collation, and indexing bug fixes: Strengthening privilege consistency, deferrable index builds, collation handling, and memory safety.
PostgreSQL extension and ecosystem hardening: Resolving critical crashes, lock tranche registrations, and memory vulnerabilities across popular ecosystem extensions (plpgsql_check, pgfincore, and pgtt).

1. Logical replication and conflict management

Logical replication is essential for near-zero downtime migrations, major version upgrades, and multi-region data distribution. Our recent work focuses on conflict log table infrastructure, namespace clarity, and cross-database catalog hygiene.

Conflict log table infrastructure

Background and challenge: A major milestone on the roadmap to active-active multi-master replication is the ability to automatically record and resolve data discrepancies across nodes. Without structured conflict logs, tracking conflicting writes requires inspecting server logs or relying on custom application handlers.
Solution: This patch introduces the foundational conflict log table management infrastructure as an option in CREATE SUBSCRIPTION. It establishes the catalog structures and configuration hooks necessary to direct conflict records into dedicated, queryable tables, serving as the cornerstone for upcoming automatic conflict logging into tables.
Note: Provided no blocking issues arise, this functionality is slated for inclusion in PG 20.
Contributors: Dilip Kumar (Primary Author)

Fix REASSIGN OWNED for subscriptions in other databases

Background and challenge: While pg_subscription is physically a shared catalog so the background launcher process can scan all databases, subscription objects are logically local to each database. Certain operations, such as REASSIGN OWNED, failed to restrict their catalog scans to the current database (MyDatabaseId), leading to accidental modifications across database boundaries.
Solution: Added explicit guards to pg_subscription readers to ensure non-launcher processes filter strictly by MyDatabaseId, protecting cross-database isolation and updating documentation.
Contributors: Dilip Kumar (Author)

Schema-qualified names in EXCEPT clause error messages

Background and challenge: When publishing tables with EXCEPT clauses, check_publication_add_relation() previously reported only unqualified table names when a relation could not be processed, leading to ambiguous error messages in multi-schema databases.
Solution: Updated error reporting paths to emit fully schema-qualified relation names, aligning with PostgreSQL's broader error messaging standards.
Contributors: Dilip Kumar (Author)

2. Core engine performance and administrative enhancements

Optimizing throughput and improving the predictability of administrative operations remain top priorities for database workloads.

Timeout handling in pg_promote()

Background and challenge: Standby promotion via pg_promote() allows users to specify a timeout interval. Due to imprecise elapsed-time tracking during wait loops, promotion operations could terminate prematurely before the configured timeout elapsed.
Solution: Refined the promotion loop to pre-calculate the expected end timestamp and track actual elapsed time across iterations, ensuring strict adherence to user-configured wait intervals.
Contributors: Robert Pang (Reporter and Author)

Sequential scan performance optimization

Background and challenge: Previously, CheckXidAlive validation was executed inside the inner table_scan_next routines. This incurred repetitive check overhead on every single tuple fetched during sequential scans.
Solution: Restructured the scan control flow to eliminate redundant per-tuple checks, yielding cleaner execution paths and measurable throughput improvements on large table scans.
Contributors: Dilip Kumar (Author)

3. Core engine access control, collation, and indexing bug fixes

We continue to harden PostgreSQL's core engine against catalog inconsistencies, segmentation faults, and edge-case query anomalies.

Large object access with pg_{read,write}_all_data

Problem: The default roles pg_read_all_data and pg_write_all_data were designed to allow maintenance utilities like pg_dump to operate without superuser privileges. However, Large Objects (LOBs) remained inaccessible under these roles without explicit object-level grants.
Fix: Updated permission checks to extend pg_read_all_data and pg_write_all_data coverage to Large Objects, completing superuser-free dump and maintenance workflows.
Contributors: Nitin Motiani (Author), Dilip Kumar (Reviewer)

Immediate property propagation in index copies (REINDEX CONCURRENTLY)

Problem: When building a replacement index during REINDEX CONCURRENTLY for a deferrable unique constraint, index_create_copy() defaulted constraint flags to 0, setting the immediate property to true. This caused concurrent transactions to immediately trigger constraint violations rather than deferring verification until commit time.
Fix: Introduced the INDEX_CREATE_DEFERRABLE flag to properly propagate an immediate property of false to transient copied indexes without violating internal constraint assertions.
Contributors: Nitin Motiani (Author)

LIKE matching with nondeterministic collations and backslashes

Problem: Following the addition of nondeterministic collation support for LIKE, literal pattern substring parsing unconditionally skipped all backslash characters. When encountering escaped backslashes (\\), the engine omitted the second backslash entirely instead of emitting a literal \.
Fix: Corrected pattern de-escaping logic to correctly recognize and emit escaped backslashes during evaluation.
Contributors: Nitin Motiani (Author)

DSM lock release and crash prevention

Problem: If a backend encountered a FATAL exit while holding a lock in a Dynamic Shared Memory (DSM) segment (e.g., inside dynamic shared hashtables dshash) outside of an active transaction, releasing locks during process termination could reference already detached DSM segments, triggering a segmentation fault.
Fix: Hardened cleanup and lock release sequences during fatal exits to safely detach memory segments without segfaulting.
Contributors: Dilip Kumar (Reviewer)

4. PostgreSQL ecosystem and extension hardening

Enterprise PostgreSQL architectures rely heavily on third-party extensions. Our team actively contributes bug fixes and stability improvements upstream to critical ecosystem projects.

plpgsql_check: LWLock tranche registration for PG14

Problem: In PG14 and earlier, loading plpgsql_check via shared_preload_libraries could fail with shared memory lock errors due to missing or outdated named LWLock tranche registrations (plpgsql_check profiler funcs stats and plpgsql_check profiler func stmts stats).
Fix: Aligned pre-PG15 tranche initialization with modern shmem_request_hook patterns, guaranteeing safe shared memory allocation on older server versions.
Contributors: Aniket Jha (Author)

pgfincore: Memory safety hardening

Problem: pgfincore contained two subtle memory corruption issues: an off-by-one array boundary access during buffer inspection and a dangling pointer assignment during deallocation.
Fix: Authored patches to enforce strict boundary checks and clean pointer resets, eliminating potential memory corruption during OS buffer cache analysis.
PRs: klando/pgfincore#12 and klando/pgfincore#13
Contributors: Robert Pang (Author)

pgtt: Use-after-free prevention on cached plans

Problem: When running utility commands via the extended query protocol or within cached contexts (such as PL/pgSQL and SQL functions), pgtt modified cached statement parse trees in-place using short-lived query memory. Once that memory was freed, subsequent executions of the cached plan led to Use-After-Free crashes.
Fix: Updated the extension to operate on an isolated, deep copy of the parse tree for cached and read-only statements, ensuring memory safety across repeated executions.
Contributors: Sunaina Punyani (Author)

Community roadmap: Your feedback matters

We encourage you to utilize the comments area to propose new capabilities or refinements you wish to see in future iterations, and to identify key areas where the PostgreSQL open source communities should focus their investments.

Acknowledgments

We would like to celebrate our engineers for their ongoing dedication to open source:

Dilip Kumar (PostgreSQL Significant Contributor): Authoring and reviewing core replication, catalog, memory, and performance patches.
Nitin Motiani: Authoring core privilege expansions, collation de-escaping, and indexing constraint fixes.
Robert Pang: Authoring promotion timing fixes and hardening pgfincore memory safety.
Aniket Jha: Hardening plpgsql_check shared memory lock mechanics.
Sunaina Punyani: Resolving memory and execution safety in pgtt.

We also extend our sincere gratitude to the wider PostgreSQL open source communities—especially the committers, reviewers, and extension maintainers—for their collaborative reviews and shared commitment to keeping PostgreSQL the world’s most advanced open source database.

Reconnecting with the heart of open source: Highlights from our 2026 GSoC India tour

2026-09-17T11:30:00.000-07:00

by Mary Radomile, Stephanie Taylor & amanda casari, OSPO

For over twenty years, Google Summer of Code (GSoC) has welcomed new developers into open source by pairing them with experienced mentors on real projects. This spirit is especially vibrant in India, which is home to more than 55% of all global GSoC participants over the last decade.

This July, our team traveled across Bengaluru and Delhi to host a series of developer events and debut our first-ever GSoC Alumni CAMP, bringing together members of India’s vibrant GSoC alumni community. We engaged directly with current and former GSoC Contributors, Mentors, and project maintainers, experiencing firsthand the passion and energy of the Indian developer ecosystem.

Community stories: Learning to think and lead as an engineer

In both Bengaluru and Delhi, a highlight of the trip was hearing directly how open source and GSoC have fundamentally changed how developers think and work.

For many attendees, having a dedicated open source mentor through GSoC took the fear out of tackling new and intimidating codebases. One former participant told us they almost walked away from a distributed storage project because it felt too overwhelming: "Storage systems felt impossible. But great mentors taught me how to think, not just how to code." Another echoed that shift in perspective: "Why am I spending so much time thinking rather than coding? Then I realized that building products is actually about thinking more than coding. GSoC taught me to think like an engineer."

That shift in mindset turns first-time contributors into long-term open source community leaders. We met one developer who submitted their very first pull request in 2023, started mentoring in 2024, and is now a lead maintainer for a major open source Android app. We also saw how new contributors to global projects can spark entire local ecosystems—like the Indian compiler community, which started with a few GSoC alumni and has rapidly grown into a 3,500+ member network with dozens of meetups across the country.

The Google Summer of Code Alumni CAMP — Bengaluru

Open source mentorship in the age of AI

Across our sessions and unconference discussions, one recurring conversation resonated above all others: the evolving role of mentorship in an AI-assisted world. The human element of open source is more critical than ever. As one attendee noted:

AI can generate the slides, but the context takes nine years.

We heard over and over from participants—open source maintainer time and attention remains a limited resource. CAMP participants presented multiple examples of where AI is proving effective for generating starter templates, writing tests, or fixing syntax. Even with this, what open source projects fundamentally need hasn't changed: maintainer time, clear architectural vision, and thoughtful code reviews.

Beyond code quality, the industry agrees that dedicated mentorship is the vital bridge between temporary contributions and long-term project stewardship. Without structured guidance, newcomers often struggle with unwritten project norms, complex codebase histories, or public review feedback, leading to contributor burnout and abandoned pull requests. Programs like GSoC transform casual interest into a sustainable maintainer pipeline by fostering psychological safety, belonging, and accountable relationships. By investing directly in maintainer time and human connection, GSoC ensures that open source projects remain secure and resilient for generations to come.

What’s next?

If our trip across Bengaluru and Delhi taught us anything, it’s that the strength of open source has always come from the communities we build together, not the volume of code any one person can ship.

As developer tools evolve, our main focus for GSoC is preserving the mentorship experience that makes the program special. Manually sifting through low-quality, automated submissions wastes maintainer time and drains the energy of volunteers who signed up to mentor new peers and colleagues. We're ready to tackle these challenges directly by optimizing our program to assist and protect our community of open source maintainers, so they can focus on leading their open source projects and helping new engineers grow. You can stay updated on GSoC’s program rules and timelines at g.co/gsoc.

To everyone who joined us in Bengaluru and Delhi—thank you for your energy, your endless inspiration, and your dedication to open source!

How much should you trust your OSS data?

2026-09-03T09:00:00.000-07:00

by Sophia Vargas, Google Open Source & Andrew Nesbitt, Ecosyste.ms

Every second, open source contribution quietly shapes the software we rely on, and yet our view of this open ecosystem is surprisingly opaque. Open source development is performed in public spaces — we can see the commits, issues and comments, the APIs and endpoints are free to use — the logs are just sitting there, so why can’t we just collect all of the data?

…Said every researcher, everywhere. However in most cases of open source related data, we are only looking at part of the whole. Why am I writing this post? Because many of us (including many business decision-makers) are too comfortable with unsubstantiated data. We’ve gotten used to it. Our models assume that it's smelly and we adjust the logic and weights to compromise. When it comes to open source, our confidence is even lower, even though our resulting decisions can directly impact individuals whom we collectively depend on.

Let’s consider one of my favorite datasets: GHarchive. Started as a hobby project in 2011, this crawler has amassed more than 15 years of event data from GitHub. While this source provides a historical record of open source development on GitHub, as a real-time or comprehensive source of metrics, it's unreliable and should not be a source for volume-based metrics.

In 2025, GHarchive captured 14% fewer events than in 2024, despite steady growth in platform adoption. Since 2025, we estimate that data retention in GHarchive has fallen to ~50% and in 2026 it may be as low as 20% for some event types (see figure below). Prior to 2025, you could make the general assumption that the majority of events would be represented in this pipeline. Since 2025, we must now assume we may be missing at least half of events and possibly more — not to mention all of the additional activity that’s left out of the event API (see GitHub’s GraphQL API.)

The crawler logic behind this dataset is simple: give me all the events from the GitHub Event stream (e.g. opening pull requests, commenting on issues etc). However, the GitHub API has limitations on the number of calls per hour as well as the number of events listed, so for days with a lot of spiky activity, the crawler will miss some. Although we never assumed that this dataset was collecting 100% of events, the current architecture is showing signs of strain. We suspect that this is due, in part, to the rate of repository growth and adoption of automated tooling on GitHub. In 2011, GitHub announced it reached 2 million public repositories, and by 2026, that figure surpassed 400 million.

I want to acknowledge that building and sharing comprehensive open datasets at scale is hard. Have you ever built a pipeline only to discover that the variables changed mid year, the payload for one output is getting truncated, all your joins broke because one side of the dataset is case sensitive … I could go on. And these examples are just ordinary data issues. Building a dataset at the scale of GitHub where “Every second, more than one new developer on average joined GitHub—over 36 million in the past year”—you start running into a new set of challenges.

My own journey with open source related data began when I repeatedly found myself questioning how much we could trust our own metrics. To expand my understanding of the nuances and the limitations of open source related datasets, I reached out to Andrew Nesbitt, who has spent years digging in data trenches for the benefit of the community. Together we converged on the following issues that we wanted to highlight for the broader community.

Assembling: Assume there will be problems

When I asked Andrew ‘can you summarize the challenges you have faced assembling comprehensive datasets?’ —“I just assume I'm going to have a terrible time anyway, so I start with my best effort and fill in the gaps”. While disappointing, this aligned with most data aggregation methods I’ve reviewed—tools such as Grimoire labs and OSS insights also require multiple processes for collection, combination and reconciliation. Even with these approaches, many sources have missing, incomplete, or inconsistent information.

One source is probably not enough. If you are considering the use of an open source project, you may want to know how many maintainers work on this project, what versions are available, what their dependencies are and any active vulnerabilities or known issues. Each of these queries requires a distinct source—the development history, the dependency graph, the CVE database, etc. Ecosyste.ms strives to pull this information together into one place, but combining data from 1000+ datasets has its own unique set of challenges.

For example, my index is probably not your index. One perennial issue is inconsistent naming conventions across sources. Beyond variable type and format, repository names, versions, packages, tags, licenses, urls, etc. tend to be unique across platforms. Some are case sensitive, there are often duplicates, and anyone can change a name at any time… I’ve been keenly following the adoption of purl and SWHID, but so far I have not found one name to rule them all.

Now we have to keep this up to date: At the moment, there is no consistent way of sharing updates across platforms. Changes to names, APIs, deletions, etc. are more often discovered by errors and breakage than by scouring release notes. To keep Ecosyste.ms up to date, Andrew has written multiple syncing processes that identify or infer updates that need to be accounted for. I asked Andrew ‘If you could ask a platform/data source to change one thing, what would it be?’, “Can I crawl an endpoint that's just NEW stuff?’

Consuming: Design your pipeline for your use case

Because of LLMs, “it's now easier for anyone to try to access and build reports”. But those building quick reports are likely not going to go through the pain of being comprehensive. This is where aggregated sources like GHarchive and Ecosyste.ms thrive. As data providers, we’d love if data consumers knew that:

How you collect data matters. If everyone wanted the same dataset, in the same format, at the same time, it would be simple. Depending on how the data is stored—centralized vs distributed and cached, relational vs graph, etc. —queries could be more efficient (in cost and computation) than exports or bulk requests faster than individual requests. This all depends on the topology of the infrastructure and the dataset. In a perfect world, data producers would design their architecture for their top user journeys. However open source related datasets serve a wide variety of user personas from corporations to non-profits, researchers to individual users, maintainers, funders, and many more, with a variety of demands from historical deep dives to realtime feedback. Data producers can’t design for all of these cases, so my challenge to them is to be more open about the best way to access this information.

At the end of the day, we have to respect the human infrastructure: Open source-related datasets are riddled with personally identifiable information (PII). Some individuals may be comfortable sharing their information with fellow contributors, but seeing it aggregated across platforms can be uncomfortable. Any source with PII should be handled with care: anonymize when you can and ensure you are in alignment with policies and regulations. Open source communities are real people so please, consume their data responsibly.

Interpreting: Never stop asking questions

While many have moved on from ‘data-driven’ to ‘AI-enabled’, the fact remains that ALL AI SYSTEMS DEPEND ON DATA. Our data about open source will continue to be incomplete and imperfect, but by asking questions about our sources, acknowledging the gaps, and considering both the technical and human processes behind open source development, we can refine and improve on how we interpret our insights and models even if they don’t completely reflect reality.

Securing the agentic era: Introducing formal verification for CEL

2026-08-18T11:30:00.000-07:00

by Sean Huh, Common Expression Language Team

We are rapidly entering an era where AI agents can autonomously draft, refactor, and deploy policies that protect our users and our systems. But this velocity introduces a vital question: How do we trust AI-generated policies?

Unit tests may fail to cover the infinite set of possible inputs that occur in production; thus, an AI agent that overfits its policy to existing tests may fail spectacularly in production. To secure automated policy authoring, we must combine heuristic testing with mathematical proofs.

We are thrilled to announce the Common Expression Language (CEL) Formal Verification Framework is now available. Powered by the Z3 theorem prover, this framework allows you to prove the correctness of your CEL expressions and policies, serving as the ultimate safety net for the agentic policy.

Automated reasoning definitively answers questions like:

“Is there any combination of inputs that allows an unapproved request into production?”
“Are we absolutely certain this AI-refactored policy matches the original behavior?”
“Can a bad actor manipulate this rule to force an evaluation error?”

Formal verification establishes mathematical certainty across the infinite spectrum of inputs. Proven policies protect your users and system while giving auditors clear proof of compliance.

To see these capabilities in action, watch our video demonstrating how the CEL Verifier REPL catches subtle logic flaws in seconds:

Proving rules from the ground up

Getting started with formal verification doesn’t require learning complex architectures right away. You can evaluate simple standalone CEL expressions to catch edge cases that tests easily miss.

(Note: The examples below use our interactive REPL syntax—check out the REPL documentation to follow along!)

1. Catching logic bugs in simple expressions (Equivalence)

How do you guarantee a refactored rule behaves identically to the original? Suppose we have a policy that allows ports 80 or 443 in production. An agent might factor the is_prod check like so:

equiv
  (is_prod && port == 80) || (is_prod && port == 443) 
  <=>
  is_prod && port == 80 || port == 443

Because logical AND has a higher operator precedence than OR, the verifier immediately flags Violated, and outputs the exact exploit: in a non-production environment (is_prod = false), the rule mistakenly allows port 443. Fixing the grouping parentheses returns Verified.

2. Enforcing exhaustive guardrails (Validity)

This capability scales directly to use cases like Kubernetes Validating Admission Policies. Suppose an engineer writes a guardrail expression that assumes every request will either be on a low port (under 80) or a high port (over 1024):

valid request.port > 1024 || request.port <= 80

When we check validity (whether an expression holds true for all inputs), the verifier exhaustively searches the entire integer space, flags Violated, and outputs the exact counterexample:

[VIOLATED] Condition is not always true. Counterexample input:
  request.port = 81

3. Guaranteeing security invariants with CEL Policy

While the verifier works perfectly with standalone CEL expressions, complex environments compose multiple rules and variables. Here, the CEL policy format shines. Using assume and assert blocks, the verifier proves a mathematical implication: if the assumptions hold, the assertions must also hold.

name: workload_admission
rule:
  variables:
    - is_admin: 'request.auth.claims.groups.exists(g, g == "admin")'
  match:
    # A subtle flaw introduced during authoring:
    - condition: 'request.is_privileged && request.is_prod'
      output: 'true'
    - condition: 'variables.is_admin || request.has_approval'
      output: 'true'
    - output: 'false'
verification:
  invariants:
    - id: universal_no_unapproved_privileged_prod
      assume:
        - 'request.has_approval == false'
        - 'variables.is_admin == false'
      assert:
        - 'rule.result == false'

The first condition admits privileged workloads into production without checking for approval or admin status. The verifier flags this and provides an example that exploits the issue:

Invariant 'universal_no_unapproved_privileged_prod' violation detected. Counterexample input:
  request.is_privileged = true
  request.is_prod = true
  request.has_approval = false
  request.auth.claims.groups = []

Assertions and assumptions define the boundaries of acceptable agent behavior, allowing developers to configure CI/CD pipelines to validate AI-generated changes simply and securely.

Under the hood: High-fidelity mathematical modeling

Translating a dynamic language into the Satisfiability Modulo Theories (SMT) domain requires immense engineering rigor to prevent the solver from hanging or hallucinating bugs. Our engine provides:

Zero false positives via three-pass taint tracking

Traditional verification tools are prone to “solver hallucinations”—reporting fake bugs when encountering custom domain-specific functions or external variables they don’t fully understand. To eliminate this noise, if a potential issue relies on an unmapped custom function, the verifier isolates and flags it as Inconclusive rather than breaking your CI pipeline with a false alarm. This guarantees every Violation report is a 100% real, reproducible bug.

Deep structural extensionality

The Formal Verification Framework offers configurable-depth bounded-model checking to prevent infinite loops within SMT quantifiers. These configurable limits allow you to control the cost of verification when analyzing deep structure equivalence in expressions like [[1], [2]] == [[1], [2]].

The mandatory bridge of trust

In the agentic era, code writes code. Mathematical proof isn’t just a nice-to-have; it is the fundamental bridge of trust developers require to let AI operate autonomously in their most sensitive systems. Get started with the CEL Formal Verification Framework, to take the next step toward a more secure agentic future today!

Let us know what you think—issues, pull requests, and feedback are always welcome!

Google joins the OpenROAD Initiative as principal member to accelerate open source silicon innovation

2026-08-11T11:30:00.000-07:00

by Ethan Mahintorabi & Aaron Cunningham, Hardware Toolchains Team

Google is committed to advancing open source silicon innovation. We are excited to share that we have formally joined the OpenROAD Initiative (ORI), Inc. as a principal member. ORI is a nonprofit public benefit corporation dedicated to the open source electronic design automation (EDA) ecosystem. As part of this commitment, Aaron Cunningham has been appointed to the ORI Governing Board to represent Google and help drive the foundation’s strategic direction, financial sustainability, and technical stewardship.

Driving long-term open source sustainability

The OpenROAD Initiative’s mission is to advance and sustain the open source EDA ecosystem by fostering collaborative innovation across research, education, and industry—transforming ideas into silicon. Google’s membership aligns directly with ORI’s multi-year sustainability goals, supported by the US National Science Foundation’s (NSF) Pathways to Enable Open-Source Ecosystems (POSE) program.

With Google’s participation and membership commitment, ORI will continue to strengthen, grow, and sustain its open source ecosystem through key vectors:

Neutral Stewardship: Fostering transparent governance where no single company has outsized control over the code, ensuring the project remains inspectable, accessible, and community-driven.
Ecosystem Growth: Supporting open and reproducible silicon research, developing robust design flows, and hosting global design contests.
Workforce Development: Supporting global silicon skilling initiatives by expanding open source chip design curricula and collaborating with academic institutions and industrial training networks.
Technical Strengthening: Enhancing continuous integration and deployment (CI/CD) pipelines, expanding PDK enablement, and improving user experience.

Leadership perspectives

“The OpenROAD Initiative is built on the vision of making chip design open and accessible to all—building a collaborative ecosystem driven by transparency and shared innovation,” said Andrew Kahng, board member of the OpenROAD Initiative. “Google’s deep commitment to open source software and hardware makes them an ideal partner. By joining at our highest membership tier, Google is helping to ensure that the open source EDA ecosystem has the stable, long-term governance and financial foundation required to grow.”

“Cutting-edge silicon research requires robust, inspectable, and reproducible toolchains,” said Drew Wingard, Director of Silicon Infrastructure, Tools and Methodology at Google. “OpenROAD has already made an incredible impact across academia and the broader industry, enabling many successful tapeouts. Google is proud to support the OpenROAD Initiative’s mission to scale this open infrastructure for the next generation of developers.”

About the OpenROAD Initiative and OpenROAD project

The OpenROAD Initiative, Inc. is a California-based 501(c)(3) nonprofit organization that provides governance, stewardship, and coordination for the OpenROAD ecosystem. The OpenROAD Project is an open source, autonomous digital chip design toolchain that democratizes semiconductor design, enabling a complete RTL-to-GDSII flow in less than 24 hours with no human in the loop. Grounded in academic research and referenced in over 500 peer-reviewed publications, OpenROAD has lowered the barriers to hardware innovation, enabling thousands of students, researchers, and startups worldwide to design and manufacture chips.

To learn more about the OpenROAD Project and install the toolchain, visit the new OpenROAD website.

For more information about membership tiers and the foundation’s governance, visit the OpenROAD Initiative website at www.openroadinitiative.org or contact membership@openroadinitiative.org.

Adapting open source practices to an AI-first world: A retrospective on 2025

2026-08-03T11:30:00.000-07:00

by Sophia Vargas, Google Open Source

Even as AI adoption accelerates and transforms the global technology landscape, open source remains foundational to how Alphabet builds, uses, and collaborates on products for billions of users. Our commitment to open source remains broad and consistent, including sharing our work year-over-year, and reflecting on what we've learned.

In 2025:

Roughly 10% of Alphabet's full-time workforce actively contributed to open source projects. This contribution ratio has remained steady over the past five years, scaling to match our growth.
These open source contributions are not just solely focused on Google. Our top projects by unique contributors at Alphabet include community-led projects such as LLVM, vLLM, Envoy, and Rust, as well as Google-initiated projects like Kubernetes, Apache Beam, and gRPC.
In addition, Alphabet projects received commits from more than 20,000 non-Alphabet affiliated user accounts.

Working together on emerging standards

Open source communities continue to provide vital collaborative spaces to define emerging standards, ensuring the interoperability and extensibility for the next generation of technologies. In 2025, we worked with more than 50 partners on the Agent2Agent (A2A) protocol to enable AI agents to communicate with each other, securely exchange information, and coordinate actions on top of various enterprise platforms and applications. Within weeks of our initial announcement, Google donated the A2A project to the Linux Foundation as part of our long-standing commitment to develop "open, collaborative ecosystem – offering greater autonomy and multiplying productivity."

Launching tools with transparency

Open source licenses provide a framework for anyone to explore, test, fork and expand on our technologies. Over the last 15 years, Google has created more than 15,000 public repositories on GitHub. Today, Google continues to maintain more than 5,000 public repositories on GitHub, and more than 1,500 public repositories on Git-on-Borg. A quick look back to some highlights from our 2025 launches includes:

Gemma 3 – a collection of lightweight open models built from the same research and technology that powers our Gemini 2.0 models. They are designed to run directly on devices — from phones and laptops to workstations — helping developers create AI applications, wherever people need them. Since Gemma’s release in 2024, the Gemmaverse community has created more than 60,000 Gemma variants. We’ve continued work on this collection and in 2026 we released Gemma 4 under an Apache 2.0 license.
Agent Development Kit – an open-source framework and SDK designed to help developers build, compose, and run both conversational and non-conversational AI agents.

Adjusting our collective security practices

As the threat landscape evolves and accelerates in the AI era, we are working on novel solutions to mitigate vulnerabilities at scale and remove some of the burden from overloaded maintainers. In 2025, we introduced several critical security initiatives to support upstream open source projects:

CodeMender: An agent designed to be both reactive, instantly patching new vulnerabilities, and proactive, rewriting and securing existing code and eliminating entire classes of vulnerabilities in the process. In its first six months, we were able to upstream 72 security fixes to open source projects, including some as large as 4.5 million lines of code.
OSS Rebuild: A security initiative to prevent software supply chain attacks by verifying build provenance. It automates package rebuilding and semantically compares results to upstream artifacts to detect tampered code on registries like PyPI, npm, and Crates.io.
OSV-Scalibr: (Software Composition Analysis Library) the core engine for vulnerability scanning used internally and in OSV-Scanner.
Credential Scanning on deps.dev: A service (using Veles) that actively scans open-source packages on deps.dev for leaked GCP credentials to prevent account compromise.

Working together to sustain the contributor community

Beyond security contributions, we remain committed to providing financial support directly to projects and maintainers. In 2025, the Open Source Programs Office (OSPO) directed $2M in sponsorships and investments to more than 40 open source projects.

Google Summer of Code celebrated its 21st year of enabling open source organizations to find, mentor, and onboard new contributors, directly supporting 1,280 individuals to contribute to 185 organizations. Over its lifetime, the global program has connected more than 23,000 participants from 125 countries with over 1,000 open source organizations globally. Moving forward, we recognize the need to evolve our program structures to ensure they continue to effectively support maintainers in the AI era.

Evolving the way we work in open spaces

Respecting community norms and practices is fundamental to establishing and maintaining trust in open source communities. To ensure we remain supportive members of your community, we encourage projects to document preferred practices and policies in the wake of new technologies. As AI changes the way we work, we are continually evaluating how best to evolve the way we engage in open spaces to preserve, prepare and bolster the communities we depend on.

We are deeply grateful for the many individuals and organizations that have worked with us to create global technologies from which everyone can benefit. You can continue to learn more about our open source initiatives, ongoing programs, and new projects at opensource.google.

Appendix: About this data

This report features metrics provided by many teams and programs across Alphabet. In regards to the code and code-adjacent activities data, we wanted to share more details about the derivation of those metrics.

Data sources: These data represent the activities of Alphabet employees on public repositories hosted on GitHub and our internal production Git service Git-on-Borg. These sources represent a subset of open source activity currently tracked by Google OSPO.
Business and personal: Activity on GitHub reflects a mixture of Alphabet projects, third-party projects, experimental efforts, and personal projects. Our metrics report on all of the above unless otherwise specified.
Alphabet contributors: Please note that unless additional detail is specified, activity counts attributed to Alphabet open source contributors will include our full-time employees as well as our extended Alphabet community (temps, vendors, contractors, and interns). In 2025, full time employees at Alphabet represented more than 95% of our open source contributors.
GitHub Accounts: For counts of GitHub accounts not affiliated with Alphabet, we cannot assume that one account is equivalent to one person, as multiple accounts could be tied to one individual or bot account.
Active counts: Where possible, we will show ‘active users’ defined by logged activity (excluding ‘WatchEvent’) within a specified timeframe (a month, year, etc.) and ‘active repositories’ and ‘active projects’ as those that have enough activity to meet our internal active-project criteria and have not been archived.

This Week in Open Source for July 16, 2026

2026-07-16T11:30:00.000-07:00

by Daryl Ducharme, Open Source Programs Office

This Week in Open Source for July 16, 2026

A look around the world of open source

We’re diving into another week of open source news and discussions. As the ecosystem continues to evolve, we're seeing more conversations around the intersection of AI workflows, security at scale, and the infrastructure that supports the open web.

This week, we’re highlighting a few "Open Source Reads" that tackle some of the biggest questions facing our ecosystem today—from the complex ethics of AI-generated workflows to the future of federated social networks. We hope these links provide valuable context as we work together to sustain the critical infrastructure we all rely on.

Upcoming Events

ASWF Open Source Days (July 19–20) — Los Angeles, CA. Hosted by the Academy Software Foundation, focusing on open source software in visual effects (VFX) and animation.
KubeCon + CloudNativeCon Japan 2026 (July 28–30) — Yokohama, Japan. The Cloud Native Computing Foundation's flagship conference in Japan.
Black Hat & DEF CON 2026 (August 1–9) — Las Vegas, NV. While primarily cybersecurity, both heavily feature open source hacking tools. The Open Source Security Foundation (OpenSSF) also has a major presence here.
FOSSY (August 6–9) — Vancouver, Canada. A highly community-focused open source conference emphasizing grassroots participation and collaboration.
Community over Code Asia (August 7–9) — Beijing, China. The Asian edition of the Apache Software Foundation's (ASF) community-first conference.
Open Source Summit Korea (August 11–12) — Seoul, South Korea. Hosted by the Linux Foundation, bringing together open source maintainers and enterprises.
KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China (September 7–9) — Shanghai, China. A massive collaborative mega-event uniting open infrastructure, cloud, and AI communities.
OSPOlogy + OSPO Summit (September 7) — Shanghai, China. Focused on Open Source Program Offices (OSPOs), governance, and corporate open source sustainability.

Open Source Reads and Links

[Blog] Elephants, Goldfish and the New Golden Age of Software Engineering - This isn't directly open source, but with the open source ecosystem trying to find good AI workflows I thought the Elephant-Goldfish model was an interesting take. It makes the distinction between using AI as a toy and using it as a tool.
[Article] Open-source AI and the Choice Before Us - The communities with minimal influence on AI development will likely be the most affected by it. This article looks at how strong rules and institutions are needed to guide AI's safe and fair use. The future of AI depends on whether we build systems that include everyone or just deepen existing divides.
[Article] Open source carries the world. Patching it at Mythos-scale can't fall to maintainers alone. - The past year has shown that AI is speeding up the discovery of vulnerabilities in open source software. Endor Labs is offering a solution to utilize AI to create patching stop gaps for users of the libraries so that maintainers and contributors have the time to create good quality fixes to these vulnerabilities.
[Blog] The AT-URI Syntax Mess - I am very excited about the open social web, especially things happening around the AT protocol. Early misunderstanding led to the current AT URI syntax being invalid. This article looks at the possible ways forward.
[Article] The Open Social Web Needs Section 230 to Survive - The open web has always been a promise of free speech. When walled garden social media got involved it created some questions around whether free speech applies to the content served by the corporations running those networks. However, the open social web is not run by corporations. It can be, and it can also be run by individuals federating instances or running their own PDS. So, how do we maintain the free speech promise of the internet with this in mind?

Which of these stories will you be chatting about at your next meetup or conference? Let us know! Share with us on our @GoogleOSS X account or our @opensource.google Bluesky account.

Google Cloud: PostgreSQL community contribution updates

2026-07-10T11:30:00.000-07:00

by Dilip Kumar, Cloud SQL for PostgreSQL & Matt Cornillon, Sales EMEA

Google Cloud is deeply committed to the long-term success of the PostgreSQL ecosystem. Our involvement goes beyond providing PostgreSQL managed services; it's also about active participation in the open source communities through technical contributions, leadership in conference committees, and sharing architectural insights that benefit all users. Following is a recap of recent events Google Cloud participated in.

PGConf.dev 2026

Serving as a vital developer-centric hub, PGConf.dev provides a unique opportunity for collaboration with the full assembly of senior PostgreSQL committers. This gathering is essential for aligning technical efforts and shaping the future project roadmap.

Key Highlights

Participation focused on strategic coordination with PostgreSQL committers regarding logical replication development, and a consultation on global index architecture.
High community interest confirms the Global Index feature solves a vital architectural requirement for enterprises.
Established community consensus to pursue a deparsing-based architectural approach for DDL replication.

Google Cloud Sessions

Dilip Kumar, a PostgreSQL contributor from Google Cloud, presenting "Experimenting with a Global Index in PostgreSQL" at pgconf.dev 2026 in Vancouver. He is speaking at a podium next to a presentation slide detailing the Global Index storage architecture and PartitionIdentifier management.

Session Title	Session Type	Speakers/Led by
Experimenting with a Global Index in PostgreSQL: Design, Implementation, and Challenges	Technical Talk	Dilip Kumar
Unconference: Global Indexes	Unconference Session	Dilip Kumar
Unconference: Logical Replication: Warts and Missing Pieces	Unconference Session	Hannu Krosing

PGConf India 2026

Key Highlights

The three-day conference was divided into a training day followed by two days of sessions. More than 580 participants attended the conference over three days.
The conference sessions included a mix of keynotes, breakout technical sessions, sponsor sessions, and booth interactions.

Google Cloud Sessions

Session Title	Session Type	Speakers
Database And GenAI	Keynote	Paresh Rathod
Experimenting with a Global Index in PostgreSQL	Technical Talk	Dilip Kumar
GCP - Best home to run PostgreSQL	Sponsor Session	Trusar Borse, Abhijeet Rajkur
Beyond shared_buffers: On-Demand Memory PostgreSQL	Technical Talk	Rajeev Rastogi, Vaibhav Popat
Where is my Memory	Technical Talk	Pushkar Kalidkar
Agentic AI Applications with GCP Databases	Keynote	Abhijeet Rajkur, Rishi Kapoor, Saurabh Gupta

PGDay Paris & PGDay France 2026

France hosts two distinct flagship PostgreSQL events, and Google Cloud is deeply embedded in both as both organizers and technical contributors. While PGDay Paris serves as an international, English-language hub for the European community, PGDay France is a community-driven, traveling event that focuses on the francophone ecosystem, taking place in Toulouse for 2026.

Key Highlights

Matt Cornillon served on the organization committee for PGDay France, while Yves Colin contributed as a member of the program committee.

Google Cloud Sessions

Session Title	Session Type	Speakers
Creating a "Dungeon Master" with Postgres and MCP	Technical Talk	Matt Cornillon
Create your first AI agent with PostgreSQL	Workshop	Matt Cornillon, Yves Colin

PGDay FOSDEM 2026

FOSDEM PGDay is a prominent open source gathering that brings together developers from across the globe to discuss the latest PostgreSQL advancements. It serves as an essential platform for exploring emerging paradigms in database development.

Key Highlights

Exploration of how AI-assisted workflows are redefining development beyond standard autocomplete for SQL queries.

Google Cloud Sessions

Session Title	Session Type	Speakers
Vibe-coding with Postgres: really?	Technical Talk	Matt Cornillon

PGConf Belgium 2026

PGConf Belgium 2026 took place at the UCLL Campus Proximus in Haasrode, Belgium, serving as an outstanding learning and networking platform for the local PostgreSQL community and students.

Key Highlights

The session was selected by faculty as supportive material for a database exam following deep student engagement.

Google Cloud Sessions

Session Title	Session Type	Speakers
Creating a "Dungeon Master" with Postgres and MCP	Technical Talk	Matt Cornillon

Nordic PG Day 2026

Nordic PG Day is the largest PostgreSQL event in the Scandinavian countries. The 2026 edition took place in Helsinki, gathering more than 130 PostgreSQL enthusiasts for a day of deep dives.

Key Highlights

Google joined as an official Partner-level sponsor for the first time, including a dedicated table booth.

Google Cloud Sessions

Session Title	Session Type	Speakers
Unlock AI Agents with PostgreSQL	Technical Talk	Mats Berglin, Miguel Toscano

Swiss PGDay 2026

Swiss PG Day is the annual event organized by the Swiss PostgreSQL User Group in Rapperswil, Switzerland. The ninth edition featured sessions in both English and German.

Key Highlights

Demonstration of the physical impact of pushing millions of vectors to PostgreSQL based on a real-world use case.

Google Cloud Sessions

Session Title	Session Type	Speakers
Surviving pgvector in production: a reality check	Technical Talk	Miguel Toscano

Postgres Conference: 2026 San Jose

Since its inception in 2007, the Postgres Conference has served as a cornerstone for advancement, fostering a rich environment for learning and professional networking.

Key Highlights

Google proudly served as a sponsor for the event.
Adapting PostgreSQL for the artificial intelligence era demands a transformation in operational approaches. With the rise of natural language tools and vibe coding speeding up development, Agentic AI places advanced demands on production databases. In their presentation, Vikas and Vishal examine how Google Cloud managed services have evolved to handle these workloads, providing architectural strategies and best practices for contemporary AI deployment.

Google Cloud Sessions

Session Title	Session Type	Speakers
Postgres and AI - Stronger Together!	Technical Talk	Vikas Arora and Vishal Bagga

Community Leadership and Committees

Googlers play a vital role in shaping the direction of the most prestigious PostgreSQL developer events. Our leadership in these committees helps ensure that enterprise-grade requirements—such as those needed for large-scale migrations—are part of the global conversation.

PGConf.dev 2026: Dilip Kumar served on the Program Committee.
PGConf India 2026: Dilip Kumar was a member of the Paper Selection Committee.
PGDay France: Matt Cornillon was a member of the organization committee and Yves Colin served as a member of the Program Committee.

Looking Forward

Our commitment remains firm: to turn feedback from these global events into code, reviews, and active community partnerships. We thank the wider PostgreSQL community and the project's committers for their continued collaboration in making PostgreSQL better for everyone.

Acknowledgement

We extend our heartfelt appreciation to our open source community contributors for their outstanding dedication and active participation in making PostgreSQL conferences a great success.

Abhijeet Rajurkar, Darshan Nagarajappa, Dilip Kumar, Hannu Krosing, Mats Berglin, Matt Cornillon, Michael Bautin, Miguel Toscano, Niranjan Shivprasad, Paresh Rathod, Rajeev Rastogi, Vaibhav Popat, Vikas Arora, and Yves Colin

Furthermore, we are deeply grateful to the broader PostgreSQL open-source communities, especially the dedicated conference organizers, committee members, and all supporting sponsors.

Community feedback: How can corporations improve support for open source maintainers?

2026-06-30T11:30:00.000-07:00

by Sophia Vargas, Google Open Source

We know that AI is actively transforming the sustainability and socio-technical dynamics of OSS communities. Google Open Source is committed to partnering with open source communities and ecosystems to learn together how we should update our own models for engagement and support.

During an open meetup for GitHub Maintainer Month, I led a session to gather community feedback on how corporations can more effectively support open source maintainers.

Paying maintainers takes creativity

Many maintainers would appreciate consistent financial support. However, facilitating payments to individuals without established contractual relationships remains a complex challenge, particularly across diverse international jurisdictions. Fiscal hosts and programs such as Open Collective, GitHub Sponsors, and the LFX Mentorship Program can simplify components of this process, but they do not resolve the underlying issues of funding sustainability and predictability. While initiatives like the Open Source Endowment are working toward long-term funding sustainability, individual maintainers also had a few ideas:

Pay per meaningful contribution vs gameable metrics: Avoid payment models based on easily manipulated units like pull request counts or review volume. A proposed alternative is ‘pay per report,' encouraging maintainers to document their achievements and upcoming roadmaps.
Commitment-based purchasing: Corporate policies might make procurement simpler (or more complex) than sponsorships, so maintainers could benefit from offering structured support services alongside traditional sponsorship opportunities.
Fund conference attendance: In-person networking can be a boon for solo maintainers but it's often cost-prohibitive. For some corporations, travel sponsorship may be a simpler alternative to direct payments.

Challenge for Corporations and Fiscal Hosts: How can we assist maintainers in understanding any and all prerequisites and documentation necessary to participate in monetary programs?

Advice from Maintainers: Consult a tax professional to understand the implications of various funding methods.

Manage and respect expectations

Beyond financial support, our discussion returned to the importance of respect and etiquette. Particularly, how can we manage expectations between heterogeneous creators, contributors and users - are maintainers clearly communicating their preferences, and are corporations actively respecting them? Some suggestions include:

Adherence to community norms: Maintainers should share their preferred communication channels, while contributors - both human and agentic - must ensure they review and follow them.
Consistency with documentation: Discrepancies between documented procedures and actual practices create friction for all participants. This standard should be upheld by both individual maintainers and corporate-managed projects.
Clarity of intent: Many maintainers would like to understand the motivation behind a contribution and reserve the right to ask questions.

To improve specific program experiences, maintainers suggested:

Consistent communication: Recipients of funding programs expect clearly communicated expectations regarding the timing and amount of disbursements.
Transparency and discoverability: Maintainers would appreciate easily discoverable records that track program participation, active agreements, and verify the status of Contributor License Agreements (CLAs).

Let's keep learning as a community

While we cannot make any promises, we want to continue to learn and challenge ourselves to consider novel ways to support OSS communities and maintainers. As a member of our community, we value your opinion. We've created a Google form to collect any thoughts you might have, as well as gauge interest in another open meeting. We plan to share any and all learnings back with the community.

Documenting the manual: how curiosity and robotic arms led to a career in open source

2026-06-22T11:30:00.000-07:00

by Daryl Ducharme, Google Open Source

When you think of "innovation" in open source, your mind probably jumps to the latest AI model or a revolutionary new framework. You might not immediately think of manual pages. Even Alejandro "Alex" Colomar, who spends his days maintaining Linux Kernel documentation, jokingly admits that some might find the work "boring" because it focuses on fixing existing issues and documenting new features rather than flashy inventions.

But as any developer knows, the most powerful code is only as good as the documentation behind it. At Google, we believe that investing in the success of projects we don't own is a core part of being a good open source citizen. That is why we are proud to sponsor Alejandro's work on the Linux Kernel man-pages project—supporting the critical infrastructure that many of our own systems rely on every day.

Documentation is the gift you give to your future self and your whole community.

The precision of a robot

Alejandro's journey into the world of essential documentation started at university. He was working with robotic arms that used a proprietary scripting language. Wanting more control, he decided to write a C library to communicate with the robots over the network by sniffing packets with Wireshark. It worked, but it was slow—he had to wait seconds between commands to ensure the robot had finished moving.

To make the movements smooth, he needed to understand the messages the robot was sending back in real-time. This required high-precision timing. He found SO_TIMESTAMP, which provided microsecond precision, but he noticed a macro called SO_TIMESTAMPNS in the header files that promised nanosecond resolution. The problem? It wasn't documented in the manual page.

The first patch

After figuring out how to use the undocumented feature by looking at the kernel source code, Alejandro decided to ensure the next person wouldn't have to struggle. He cloned the man-pages repository, wrote a new paragraph based on existing features, and figured out how to send a plain-text patch via email.

"As it was my first patch, I was a bit intimidated by the procedure," Alejandro recalls. That intimidation led to a commit message he is still proud of today: roughly 120 lines of explanation for just 25 lines of new documentation. He wanted to prove that he had done his homework. The welcoming response from the maintainer encouraged him to keep going, leading to more patches and, eventually, a career-long dedication to clarity in open source communities.

Sustaining the commons

Google understands that open source is a "small community built on trust." By supporting maintainers like Alejandro, we help ensure that critical infrastructure—like the documentation that powers the Linux ecosystem—remains accurate and accessible for everyone. We believe that using open source comes with a responsibility to contribute and sustain it, which is why we partner with developers to maintain and grow critical projects.

Alejandro's work doesn't just help himself; it helps thousands of other programmers who rely on correct documentation to build the next generation of technology. As he puts it: "I couldn't program without correct documentation, so whenever I find an issue in documentation, I try to fix it."

A garden that needs tending

We often say that a community is a garden, not a building—it requires constant tending, not just initial construction. By sponsoring Alejandro, we are helping to tend that garden, ensuring the "manual" remains a living, breathing resource for the global developer ecosystem. Whether it is fixing a typo or documenting a high-precision networking macro, every contribution makes the "eyes" on the code that much sharper.

In-place pod restarts: Boosting efficiency and workload reliability in Kubernetes v1.35

2026-06-18T11:30:00.000-07:00

by Duncan Campbell & Giuseppe Tinti Tomio, Kubernetes

Operational efficiency and system resilience are critical when running scaled platforms. Yet, in Kubernetes, recovering from software crashes remains a headache because you couldn't trigger a clean restart of a Pod's containers without recreating the entire Pod object, leading to some amount of resource waste.
To address this, Restart All Containers on Container Exits graduated to beta and is enabled by default in Kubernetes v1.36. Developed in close collaboration with the CNCF community, this capability represents Google's commitment to investing in the success of foundation-led open source projects. By sharing best practices from running large distributed systems internally, we are helping build a more resilient and efficient ecosystem. Letting containers restart while keeping the Pod's runtime identity provides a built-in way to perform in-place Pod recovery, boosting application reliability and saving resource costs.

The Problem: The High Cost of Pod Re-creation

Historically, Kubernetes managed failures using pod level restart policies. While sufficient for simple services, modern multi-container Pods often have complex dependencies. When a failure requires a full environment reset, your only option was deleting and recreating the entire Pod.
This introduces massive control plane churn, causing latency and pressure on the etcd backend during large failures:

Initialization Dependencies: If a main container corrupts a local environment, for example, single-use secrets that must be re-requested, restarting just that container is insufficient; the setup must run again.
Watcher Interoperability: If a watcher sidecar detects a fatal error, it must trigger a full recreate of the entire pod and its infrastructure, including the sandbox.
Stale States: If a database sidecar proxy restarts, the main application can get stuck attempting to use stale, broken connections.
Resource Race Conditions: When a large job finds a proper set of nodes, recreating Pods can lead to other pending Pods taking over those resources. In-place restarts eliminate this race condition risk.

Previously, resolving these failures required destroying the entire Pod. For large batch or AI/ML workloads, where thousands of Pods might fail simultaneously, this can lead to "Thundering Herd" scheduling requests, delaying recovery and wasting expensive GPU/TPU compute time.

Introducing In-Place Restarts: The RestartAllContainers Action

Kubernetes v1.35 introduces the RestartAllContainers action, enabled by the RestartAllContainersOnContainerExits feature gate, which graduated to beta in 1.36 alongside its dependencies ContainerRestartRules and NodeDeclaredFeatures. This lets a container's exit behavior trigger a fast, in-place restart of the entire Pod on its existing node.
The Kubelet halts all containers while keeping the Pod sandbox intact, preserving critical infrastructure:

Network Identity: Keeps the same IP, network namespace, and UID, completely bypassing IP reassignment.
Hardware and Devices: Keeps GPUs/TPUs bound, eliminating scheduling and re-allocation delays.
Storage Mounts: Volumes, including emptyDir and PVCs, remain fully mounted; their content is not cleared during restarts.

Once terminated, the Kubelet re-runs init containers (including sidecars, which are part of the init sequence) in order, guaranteeing a clean setup in a known-good environment.

A Native Pod Specification Example

You can implement this under the container's restartPolicyRules field. Here is a quick example of how a watcher sidecar can trigger an in-place restart of the entire Pod by exiting with code 88:
YAML
Note: Image names and paths in the YAML below are for illustrative purposes.

apiVersion: v1
kind: Pod
metadata:
  name: ml-worker-pod
spec:
  restartPolicy: Never
  initContainers:
    - name: setup-environment
      image: registry.k8s.io/ml-tools/setup-worker:v1.0
    - name: watcher-sidecar
      image: registry.k8s.io/ml-tools/watcher:v1.0
      restartPolicy: Always
      restartPolicyRules:
        - action: RestartAllContainers
          exitCodes:
            operator: In
            values: [88]
  containers:
    - name: main-application
      image: registry.k8s.io/ml-tools/training-app:v1.0

The Operational Impact of In-Place Restarts

For organizations running distributed workloads, RestartAllContainers provides serious operational advantages:

No Control Plane Overhead: By preserving identity, clusters avoid scheduling latency and DNS propagation. This was a key factor for JobSet using this feature to reduce recovery from minutes to seconds.
Node Locality Preservation: Since the Pod stays anchored to the same node, restarted containers can instantly access local, warm storage caches.
Maximized Hardware Efficiency: In distributed AI training, losing a single node halts the entire job. Keeping accelerators like GPUs/TPUs bound lets workloads resume training significantly faster, directly reducing compute costs.

Observability and SRE Best Practices

To support monitoring, Kubernetes v1.35 introduces the AllContainersRestarting Pod condition. Set to True during restarts, it alerts SREs and autoscalers, preventing false-positive alerts, while container restart counts increment to let Prometheus easily track recovery events.
To use in-place restarts successfully, shift your mental model to "persistent sandboxes" and follow three best practices:

Ensure Reentrancy: Kubelet only guarantees "at least once" execution for init containers. Reentrancy is now a standard requirement, so your code must be fully idempotent.
Plan for Termination Handling: Graceful termination (preStop hooks) is not supported for in-place restarts. SIGKILL is almost immediate, so applications must handle sudden exits gracefully.
Prepare External Tooling: CD and observability tools should expect re-running init containers without interpreting them as new deployments.

What's Next?

This beta capability is a major step toward fluid workload management and serves as a building block for advanced community features like JobSet in-place restarts (KEP-467).
Our work on KEP-5532 reflects our commitment to transparent open source governance. Developed collaboratively within SIG Node, this feature shows how we hold ourselves to high citizenship standards; making our design, goals, and intentions transparent while building shared best practices that benefit everyone. We encourage you to experiment with Kubernetes v1.35 and share your feedback with the community!

Learn More

Read the Kubernetes Pod Lifecycle Documentation.
Explore KEP-5532: Restart All Containers on Container Exits.
Review theKubernetes v1.35: Restart All Containers Blog Post.
Join the SIG Node Community on Slack (#sig-node).

Open rails for agentic commerce at Open Source Summit North America 2026

2026-06-16T16:16:03.340-07:00

by Anurag Sinha, Universal Commerce Protocol (UCP)

At Open Source Summit North America 2026, I shared why agentic commerce needs open rails.

As AI agents become more capable, the shopping journey is shifting from "show me" to "help me." Instead of browsing, comparing, clicking, and checking out step by step, people can increasingly ask an agent to help them decide what to buy and, in some cases, complete the purchase. Industry forecasts suggest agentic shopping could account for roughly 10% to 25% of U.S. e-commerce by 2030 (Bain), which points to a meaningful shift in how digital commerce will work. Watch the full keynote here.

Why shared rules matter

That shift also exposes a challenge. Commerce is still highly fragmented. Different businesses, payment providers, and platforms operate with their own rules, workflows, and business logic. Every new surface adds more integration work. Every bespoke connection creates more complexity. And that fragmentation makes it harder for AI systems to understand and perform commerce actions consistently across businesses. A shared language lowers that barrier for everyone.

A common language for agentic commerce

That is the problem Universal Commerce Protocol (UCP) is designed to solve.

We launched the Universal Commerce Protocol, or UCP, with industry leaders to establish an open standard for agentic commerce, built to work across the shopping journey. UCP creates a common language for agents and systems to operate together across consumer surfaces, businesses, and payment providers, so the ecosystem does not need a different bespoke integration for every new agent or platform.

Just as importantly, UCP is designed for the real world. Every business has its own way of selling. Checkout, fulfillment, loyalty, policy logic, shipping, and post-purchase flows can vary widely between a local shop, a marketplace, and a large retailer. UCP is built to support that reality.

A layered architecture for a shared commerce language

UCP uses a layered model to create a reusable shared language for commerce. Services organize domains like shopping and common. Capabilities define core actions such as checkout, catalog, cart, orders, and shared functions like identity linking. Extensions keep those capabilities configurable, so features like fulfillment can be modeled once and reused across multiple flows instead of being hardwired each time. At the transport layer, UCP stays agnostic, supporting bindings like REST, Model Context Protocol, and Agent2Agent.

Together with capability discovery and payment handling, these layers help consumer platforms, agents, and businesses interoperate more consistently over time. They also let different participants advertise what they support, compose new behaviors, and communicate over the transport that works best for them.

Built in the open

A standard for everyone should be shaped by everyone. Because UCP is open, merchants, developers, and community contributors can pressure-test real-world gaps, propose new capabilities and extensions, and help make sure the protocol reflects more than the needs of the largest players. That kind of participation is what keeps an ecosystem moving.

Since launch, UCP has continued to evolve through new capabilities, an expanded Tech Council, and new consumer experiences built on top of the protocol. That momentum matters because standards only work when the ecosystem uses them.

Watch the full keynote

Agentic commerce is still evolving, and UCP is a foundational building block to support what's next in this new era.

If you want the full architecture walkthrough and the complete story from Open Source Summit North America, watch the session here. And if you want to go deeper, you can explore the UCP documentation, join the community conversation, and contribute to the public repository.

CEL finds a new home at github.com/cel-expr!

2026-06-16T11:30:00.000-07:00

by Olena Huang, CEL (Common Expression Language) team

We're excited to announce that the official Common Expression Language (CEL) repositories have moved to a dedicated GitHub organization. Visit the new cel-expr repository now!

Why the move?

This move is a key step in strengthening the CEL ecosystem. By centralizing our projects, including the language specification, Go, C++, C, Java, and Python implementations, under the cel-expr organization, we aim to:

Enhance Branding: Create a clear and unified brand identity for CEL.
Improve Discoverability: Make it easier for users and contributors to find all official CEL resources in one place.
Ensure Consistency: Foster consistency across all CEL projects.
Streamline Development: Simplify our development and release processes.

What's Changing?

The following repositories now reside in the cel-expr organization:

google/cel-spec is now cel-expr/cel-spec
google/cel-cpp is now cel-expr/cel-cpp
google/cel-go is now cel-expr/cel-go
google/cel-java is now cel-expr/cel-java
cel-expr/cel-python and cel-expr/cel-c have already been in the cel-expr namespace

All future development, issues, and pull requests for these projects will take place in their new homes within the cel-expr organization. This is a non-breaking change, due to automatic redirects, but you should update your URLs where possible.

What Stays the Same?

We've worked to make this transition as seamless as possible:

Automatic Redirects: GitHub will automatically redirect all web traffic and git operations from the old google/cel-* URLs to the new cel-expr/cel-* locations. Your existing links and git remote configurations pointing to the old URLs should continue to work for cloning and fetching.
Preserved History: The full commit history, issues, and pull requests for each repository have been migrated and are available in the new locations.

Action Required: Update Your Dependencies

While existing links and git remote configurations pointing to the old URLs should continue to work thanks to GitHub's redirects, we recommend updating your dependency management configurations (e.g., go.mod, pom.xml, requirements.txt, etc.) to point directly to the new repository URLs under https://github.com/cel-expr. This ensures you are fetching the latest code and releases from the canonical source.

We're thrilled about this new chapter for CEL, bringing all our core components under one roof. We believe this will foster a stronger CEL community and accelerate the development and adoption of CEL.

Explore the new organization at https://github.com/cel-expr!

A new pkg.go.dev API for Go

2026-06-12T11:30:00.000-07:00

by Ethan Lee, Jonathan Amsterdam & Hana Kim, Go Team

Access to Go metadata has been an everpresent need for the Go community. Since its launch, pkg.go.dev has served as a central hub for Go package documentation and discovery. While we initially prioritized providing this comprehensive access via a web interface, the need for streamlined programmatic access has become increasingly clear.

Structured API access has been one of the most highly requested features for pkg.go.dev for a while now. Developers building tools, IDE integrations, automated workflows, and other systems have had to rely on inconsistent and fragile scraping methods. By providing a formal API, we can provide fast and efficient access to required data. This foundation also sets Go up for the future of AI-assisted coding. Large language models and agents can access the context necessary to reason about the Go ecosystem with greater precision and accuracy.

Empowering Tool Builders

Our goal with this API is to reduce the technical churn for builders and innovators. By offering structured JSON metadata, we address the following use cases:

Search and Discovery: The API enables fast and efficient search across the entire Go module ecosystem.
Driving AI Innovation: As AI-assisted coding evolves, LLMs and agents need precise context. This API provides the data required for agents and models to reason deterministically about Go packages.

The Service Interface

Built for stability and efficient caching, the API uses a stateless, GET-only architecture. Primary endpoints are currently hosted under the v1beta path. Following a period of feedback from the Go community and confirmed stability, we intend to transition toward a formal v1 release.

For a complete interactive reference of all endpoints, query parameters, and response shapes, see pkg.go.dev/api. The machine-readable API contract is also published directly at pkg.go.dev/v1beta/openapi.yaml.

Endpoint	Description
`/v1beta/imported-by/{path}`	Paths of packages importing the package at `{path}`.
`/v1beta/module/{path}`	Information about the module at `{path}`.
`/v1beta/package/{path}`	Information about the package at `{path}`.
`/v1beta/packages/{path}`	Information about packages of the module at `{path}`.
`/v1beta/search/search?q={query}`	Search results for a given query.
`/v1beta/symbols/{path}`	List of symbols declared by the package at `{path}`.
`/v1beta/versions/{path}`	Versions of the module at `{path}`.
`/v1beta/vulns/{path}`	Vulnerabilities of the module or package at `{path}`.

An example of retrieving package information is shown below:

curl https://pkg.go.dev/v1beta/package/github.com/google/go-cmp/cmp | jq
{
  "modulePath": "github.com/google/go-cmp",
  "version": "v0.7.0",
  "isLatest": true,
  "isStandardLibrary": false,
  "goos": "all",
  "goarch": "all",
  "path": "github.com/google/go-cmp/cmp",
  "name": "cmp",
  "synopsis": "Package cmp determines equality of values.",
  "isRedistributable": true
}

A Reference Implementation

To demonstrate how to interact with our API, we are providing a reference CLI implementation: pkgsite-cli. This implementation serves as a practical example for developers looking to build their own integrations, showing how to handle the data directly from the terminal. Note, as the API continues to evolve, the interface and behavior of this CLI may change.

You can use it to search for packages or inspect symbols without leaving your shell:

go install golang.org/x/pkgsite/cmd/internal/pkgsite-cli@latest

pkgsite-cli search "uuid"
github.com/google/uuid
  Module:   github.com/google/uuid@v1.6.0
  Synopsis: Package uuid generates and inspects UUIDs.
... more


pkgsite-cli package github.com/google/go-cmp/cmp
github.com/google/go-cmp/cmp
  Name:      cmp
  Module:    github.com/google/go-cmp
  Version:   v0.7.0 (latest)
  Synopsis:  Package cmp determines equality of values.

pkgsite-cli package --symbols github.com/google/go-cmp/cmp
github.com/google/go-cmp/cmp
  Name:     cmp
  Module:   github.com/google/go-cmp
  Version:  v0.7.0 (latest)
  Synopsis: Package cmp determines equality of values.

Symbols:
  type Indirect struct{}
  type MapIndex struct{}
  type Option interface{}
  ... more

Looking Ahead

While we prioritize stability for our new /v1beta endpoints, we are eager to hear how open source communities use these resources to solve real-world problems.

We look forward to your feedback via our issue tracker and to seeing the tools you’ll build next.

Introducing OpenRL: A self-hosted post-training API for fine-tuning LLMs

2026-06-11T11:30:00.000-07:00

by Sunil Arora, Shuby Mishra & Chuang Wang, GKE

We are pleased to share a research preview of OpenRL, a new open-source project coming out of GKE Labs. OpenRL is a self-hosted training API for fine-tuning LLMs on your own Kubernetes cluster.

Why we built it

If you look at agentic RL on LLMs, it is incredibly easy to get bogged down in system complexity. To run a single RL loop, you have to coordinate a dozen different things: selecting and cleaning datasets, choosing RL environments, debugging training loops, managing reward signals, handling inference mismatches, allocating hardware, and managing infrastructure. Picture looks something like this:

Figure shows an AI researcher and an infrastructure engineer staring at the hurdles in post training along the way to the summit.

Each of these is a hard problem. But what makes it more complex is how tightly AI research and infrastructure concerns are mixed together in today's tooling and frameworks.

We believe decoupling the infrastructure from AI research can make these problems more tractable so that infrastructure engineers and AI researchers can independently tackle them. We have seen this pattern with Kubernetes where Kubernetes abstracted out the infrastructure and made application developers and SREs life easier.

So, can you abstract out post training infrastructure? We believe so and drew huge inspiration/validation from Tinker (from Thinking Machines). The Tinker APIs for post training hit that Goldilocks zone where it hides all the post training infrastructure behind four key APIs:

Figure shows high level components and their interaction in a OpenRL based RL workflow

So the end result of this abstraction is that AI Researchers get full flexibility on their RL loop and infrastructure engineers can focus on scaling, orchestration, and reliability. OpenRL allows you to run the same training APIs but on your own infrastructure. And this decoupling has other interesting benefits.

Sharing GPUs

Traditional RL loops are strictly sequential. The trainer waits for the sampler to finish rollouts, the sampler waits for the environment to score rewards (which is often bound by slow CPU/network tasks), and the whole loop sits blocked. Your expensive GPUs spend a lot of time doing nothing. The abstraction allows running multiple RL jobs and allows infrastructure engineers to pack the training/sampling steps to utilize more of their GPUs. The graph below shows the GPU consumption in OpenRL for running one, two, and three RL jobs concurrently.

The figure shows the trainer/sampler duty cycle in OpenRL for scenarios with 1 RL job, 2RL jobs and 3 RL jobs respectively.

Better UX

Once you separate out the infrastructure behind the APIs, you start to see the gains in user experience of developing the RL loop because AI researchers no longer have to wrangle the complex python dependencies like cuda. When you are doing R&D, you do not have to run the RL loop directly on the machines with GPUs, you can simply run your RL loop on your Mac pointing to the training APIs running on a Kubernetes cluster/VMs.

Autoresearch

We believe that frontier AI research will get more and more automated in the future and abstracting out infrastructure as a building block is key to that. To demonstrate that, we added an autoresearch recipe inspired heavily by karpathy's work. The recipe demonstrates how to conduct parallel experiments to conduct parameter sweep, and improve the reward signal for our text-to-sql recipe for Gemma models.

Figure showing autoresearch UI with multiple AI researchers conducting experiments in parallel in OpenRL

What OpenRL is not

A managed service. OpenRL is self-hosted and not a managed service. We aim to make it easy for users to deploy and operate it on their Kubernetes clusters.
An RL framework. OpenRL gives AI researchers full control over their RL loop.

Get started

We have made it easy to run OpenRL on your Mac, Nvidia GPUs, or on GKE. This allows you to test your RL loop on Mac and when you are ready to scale, you can point the RL loop to the OpenRL endpoint running in the GKE cluster.

Try out our text-to-SQL example for teaching the latest Gemma model SQL here: guides.

One of the benefits of a Tinker compatible endpoint is that you can use Tinker-Cookbook with OpenRL. Tinker-cookbook is one of the best resources for post training infrastructure for RL.

Future steps

We have started with a simple architecture focussing on LoRA fine-tuning and plan to evolve the project in the coming months, so please give it a try and share your feedback. A few things we are very excited to work on:

Full parameter fine-tuning
Multitenancy (simultaneous RL on different types of base models)

Acknowledgement

We have been inspired by the work done by various open source projects in AI communities, so huge thank you to Thinking Machines, vLLM, PyTorch, prime-rl, verl, SkyRL, and llm-d.

Google Open Source Blog

This Week in Open Source for October 23, 2026

A look around the world of open source

Upcoming Events

🗓️ October 2026

🗓️ November 2026

🗓️ December 2026

Open Source Reads and Links

Introducing the Elephant-Goldfish Model plugin: scaling design-driven AI software development

Shifting left in the AI coding era

Understanding the Elephant-Goldfish Model (EGM)

Announcing the open source Elephant-Goldfish Model plugin

The EGM plugin in a nutshell

Phase 1: Growing the Elephant (egm-brainstorm)

Phase 2: Drafting with the Elephant (egm-draft)

Phase 3: The Goldfish Protocol (egm-review)

Human judgment is the artifact

Announcing the verifiable quantum circuit compilation competition on Kaggle

Turning compilation into an AI agent loop

How the competition works

Get involved

This Week in Open Source for October 15, 2026

This Week in Open Source for October 15, 2026

Upcoming Events

🗓️ October 2026

🗓️ November 2026

🗓️ December 2026

Open Source Reads and Links

How much does it cost to use open source software?

Customization incurs additional cost

Study limitations and future work

Adopting Pyrefly as the Python type checker at Google

Moving away from Pytype

Why Pyrefly?

Performance

Typing spec conformance

Actionable error diagnostics

Performance and infrastructure impact

Developer feedback

Looking forward

This Week in Open Source for October 2, 2026

This Week in Open Source for October 2, 2026

Upcoming Events

🗓️ October 2026

🗓️ November 2026

Open Source Reads and Links

Standardizing fine-grained access control in Apache Iceberg's REST catalog

The problem: there is no server in the read path

The mechanism

The nine masking actions

Fail-closed by design

The trust model, stated plainly

What is still missing

Why this matters

New public datasets available in Google Cloud Lakehouse

Prerequisites

Explore metadata with PyIceberg

PySpark and Managed Service for Apache Spark

Start building today

This Week in Open Source for September 24, 2026

This Week in Open Source for September 24, 2026

Upcoming Events

🗓️ October 2026

🗓️ November 2026

Open Source Reads and Links

Google Cloud: Investing in the future of PostgreSQL — 2026 highlights

Technical contributions: January 2026 – September 2026

1. Logical replication and conflict management

2. Core engine performance and administrative enhancements

3. Core engine access control, collation, and indexing bug fixes

4. PostgreSQL ecosystem and extension hardening

Related reading

Community roadmap: Your feedback matters

Acknowledgments

Reconnecting with the heart of open source: Highlights from our 2026 GSoC India tour

Community stories: Learning to think and lead as an engineer

Open source mentorship in the age of AI

What’s next?

How much should you trust your OSS data?

Securing the agentic era: Introducing formal verification for CEL

Phase 1: Growing the Elephant (`egm-brainstorm`)

Phase 2: Drafting with the Elephant (`egm-draft`)

Phase 3: The Goldfish Protocol (`egm-review`)