Google's open protocol for AI agents lands at the Linux Foundation
Plus: Google open-sources command-line coding agent; LanceDB lands big bucks to build the ‘open source standard for AI data’; and more.
Hi folks,
Google features prominently in this week’s edition of Forkable. The internet giant donated its open source AI agent protocol to the Linux Foundation, launched an open source coding agent for the command-line, and open-sourced a generative music model that allows users to create, control and perform music in real-time using text prompts, audio samples, or a combination thereof
Elsewhere, LanceDB — a four-year-old open source startup I recently profiled in Forkable — raised a chunky round of funding. And there’s much more… including the first ever IRL meeting between old foes Linus Torvalds and Bill Gates. Yup, they’d genuinely never met before until this month.
As usual, feel free to reach out to me with any questions, tips, or suggestions: forkable[at]pm.me.
Paul
Open issues
Google gives open source AI protocol to Linux Foundation
Some two months after unveiling its open source Agent2Agent (A2A) protocol to power AI agent communication and collaboration, Google revealed that it’s donating the project to the Linux Foundation.
While Anthropic’s MCP (Model Context Protocol) is all about connecting large language models (LLMs) and agents to external tools, systems, and data, A2A is more concerned with agent-to-agent orchestration — facilitating collaboration between different agents that specialize in different tasks. They may also need to use MCP to pull data in from elsewhere, so there is more of a complementary than competitive thing going on here.
As an official Linux Foundation project, A2A will stand a greater chance of becoming an established, vendor-neutral standard for AI agent interoperability. Indeed, the A2A project already counts a host of big-name Google rivals as collaborators under the auspices of the Linux Foundation, including Amazon Web Services and Microsoft, as well as Cisco, Salesforce, SAP, and ServiceNow.
“By joining the Linux Foundation, A2A is ensuring the long-term neutrality, collaboration and governance that will unlock the next era of agent-to-agent powered productivity,” the Linux Foundation’s executive director Jim Zemlin said.
Read more: Google donates the Agent2Agent protocol to the Linux Foundation [The New Stack]
Google’s command-line coding agent
Developers can access and interact with their code through GUIs like web browsers and desktop IDEs. But if they want more flexibility, granularity, and control, they might prefer the command-line interface (CLI). And this is why Google this week debuted Gemini CLI, an open source AI coding agent that’s accessible directly through the terminal.
Gemini CLI serves direct access to Google’s Gemini AI models and its coding assistant, Gemini Code Assist. This allows developers to ask for help understanding a piece of code, run a unit test, debug, or update their documentation — all with natural-language prompts.
Google is keen to stress the open source nature of Gemini CLI, too, for anyone who might be wary of giving Big Tech access to their codebase. The company writes:
Because Gemini CLI is fully open source (Apache 2.0), developers can inspect the code to understand how it works and verify its security implications. We fully expect (and welcome!) a global community of developers to contribute to this project by reporting bugs, suggesting features, continuously improving security practices and submitting code improvements.
Read more: Gemini CLI: your open-source AI agent [Google blog]
LanceDB lands $30M to become the ‘open source standard for AI data’
Regular Forkable readers may remember my COSS Corner profile piece last month on LanceDB, a four-year-old startup leaning on open source to build a database for multimodal AI.
Fast forward to this week, and the Y Combinator (YC) alum announced a whopping $30 million Series A round of funding, just a year after closing its $8 million seed round.
LanceDB is the handiwork of CEO Chang She, one of the original authors of Python-centric open source data analysis and manipulation library Pandas; and CTO Lei Xu (pictured together above), a former colleague at Cloudera which had acquired She’s previous startup.
By way of a brief recap, LanceDB is building a “data lake for multimodal AI,” making it easier for developers to train, pre-process, and explore all their AI data — including text, images, audio, video, and vector embeddings — in a single place.
The platform is built atop a columnar open source data format of the company’s own creation called Lance, which is 100 times faster than Apache Parquet, according to She.
On top of Lance, the company has also developed the open source LanceDB vector database. This is meant more for smaller-scale experimentation, though, and when the user is edging toward production use-cases, they will likely want to upgrade to LanceDB’s Cloud or Enterprise incarnations which offer all the automation, security, hosting, and operational expertise a company might need.
LanceDB has already amassed a fairly impressive roster of users including Midjourney and TikTok-owner ByteDance, which is using Lance for its cloud computing and AI unit Volcano Engine.
In the same way as Parquet has emerged as the standard for business intelligence and analytical data, and Apache Arrow for in-memory data, She reckons Lance is fast-becoming the standard format for storing and processing multimodal AI data.
“Four years ago my co-founder Lei and I asked a simple question: Why is working with embeddings, images, and video still so difficult, when compared to tabular data? The truth was simple: the industry was building on foundations meant for yesterday’s data,” She wrote in a blog post announcing the company’s new funding. “Since we started LanceDB, AI’s rise has been nothing short of meteoric. In two decades of building data tools, I’ve never seen technology spread faster. The generation and use of multimodal data is growing at an unprecedented pace, driven by the need to capture the full richness of meaning.”
LanceDB’s Series A round was led by Theory Ventures, with participation from CRV, YC, Databricks Ventures, RunwayML, Zero Prime, Swift, among others.
Read more: Announcing: $30m Series A [LanceDB blog]
Patch notes
An AI engine for live music
Magenta, the open source research project that emerged out of Google’s TensorFlow back in 2016, has spawned a new AI model for composing and performing music live, and interactively.
Now part of Google DeepMind, Magenta is all about showcasing how machine learning can help create music and art, with TensorFlow serving as the core framework that handles model training, inference, and deployment.
Magenta RealTime (RT), as its new model is called, is an open source, open weights generative music model that allows users to create, control and perform music in real-time using text prompts, audio samples, or a combination thereof. Magenta RT is basically the open, self-hostable alternative to Lyria RealTime which Google debuted last month, and which powers applications such as MusicFX DJ.
Read more: Magenta RealTime: An open-weights live music model [Magenta blog]
Open standards for AI in finance
A consortium of financial institutions and tech giants have thrown their weight behind a new initiative from the Fintech Open Source Foundation (FINOS) to standardize controls for AI in the financial services sector.
Common Controls for AI Services, as the project is called, is setting out to establish a shared, open source, technology‑agnostic framework that unites financial institutions and technology vendors under common, auditable standards, enabling them to deploy AI systems in a secure, compliant, and scalable manner.
The financial companies to sign up so far include BMO, Citi, Morgan Stanley, and RBC, while the tech firms include Microsoft, Google Cloud, and Amazon Web Services — this is the clearest sign that the standards will be cloud-agnostic. Other consultants and vendors on board include Red Hat, Sonatype, ControlPlane, Scott Logic, and Tetrate.
“As AI becomes increasingly integrated into financial services, establishing common, open standards defined in collaboration with our customers is essential to ensuring trust, security, and regulatory compliance as part of the shared responsibility model,” noted Allison Nachtigal, VP of Azure and chief product officer at Microsoft.
Read more: Global financial institutions and technology leaders collaborate under FINOS to launch open source common controls for AI services [Press release]
The Bryce is right
OpenStack founder and long-serving OpenInfra Foundation chief Jonathan Bryce is now also executive director for cloud and infrastructure at the Linux Foundation, in addition to a new role as executive director for the Cloud Native Computing Foundation (CNCF).
The announcement comes just a couple of weeks after the OpenInfra Foundation formally became part of the Linux Foundation.
Additionally, CNCF chief technology officer (CTO) Chris Aniszczyk will now serve as CTO for cloud and infrastructure at the Linux Foundation
So essentially, Bryce and Aniszczyk will assume more holistic roles across a broad gamut of Linux Foundation projects.
“Their combined leadership will instill new momentum across cloud native, infrastructure, and AI projects at the Linux Foundation, and ensure our communities and members continue to benefit from the most critical open source technologies on the planet,” the Linux Foundation’s executive director Jim Zemlin said.
Read more: Linux Foundation appoints Jonathan Bryce as Executive Director, Cloud & Infrastructure and Chris Aniszczyk as CTO, Cloud & Infrastructure [Press release]
Unsung hero
HeroDevs, a startup that provides companies with on-going support for deprecated open source software they rely on, announced a $20 million Open Source Sustainability Fund this week.
Unlike similar funds that target under-resourced open source project maintainers, HeroDevs’ fund is designed squarely for projects that are approaching end-of-life (i.e. the maintainers are preparing to sunset a specific version of an open source product), or where a community has stepped in to provide critical support for security patches, bug fixes, or compatibility updates.
"Open source is the foundation of virtually every application and service we use today—it runs the world. It deserves the same long-term investment, reliability, and accountability as any other enterprise infrastructure," HeroDevs CEO Aaron Frost said. “Open source creators shouldn't have to choose between their life and their legacy.”
Read more: HeroDevs launches $20M sustainability fund for open source creators to secure end-of-life software [HeroDevs blog]
And finally…
Linus and Bill finally meet
Bill Gates co-founded Microsoft 50 years ago, while Linus Torvalds created Linux nearly 34 years ago. Despite their divergent technological ideologies, you would’ve thought that the paths of these two old foes would’ve crossed by now. But it seems not — until last week, that is, at a dinner hosted by Microsoft Azure CTO Mark Russinovich.
Russinovich took to LinkedIn to share a few scant details of the meeting, which also included renowned U.S. software engineer David Cutler.
Russinovich wrote:
I had the thrill of a lifetime, hosting dinner for Bill Gates, Linus Torvalds and David Cutler. Linus had never met Bill, and Dave had never met Linus. No major kernel decisions were made, but maybe next dinner
Steven Vaughan-Nichols over at ZDNET procured some more details of the meeting via an email exchange with Torvalds himself, who noted that his chat with Gates was “almost entirely unrelated to any operating systems or software engineering.”
However, guitars were part of the discussions, apparently, while Torvalds also affirmed that the Linux and Microsoft rivalry is very much a thing of the past.
Read more: What tech titans Linus Torvalds and Bill Gates talked about in their first meeting [ZDNET]