cloudparts

LF AI & Data Foundation

The Linux Foundation’s fund for open-source AI and data projects, from ONNX, Milvus and OpenLineage to Delta Lake, Unity Catalog and Vortex, each at a sandbox, incubation or graduated stage.

What it is

The LF AI & Data Foundation is part of the non-profit Linux Foundation. Formally it is a “directed fund”: its charter, effective October 6, 2020, gives it the job to “raise, budget and spend funds” for artificial intelligence, machine learning and data-related open-source projects. It began as LF Deep Learning on March 26, 2018, with one project, Acumos AI; it was renamed LF AI in May 2019 and merged with ODPi, a consortium formed around Hadoop and big data, in October 2020. In October 2026 its projects page sorted 72 projects by stage: 12 graduated, 23 in incubation, 18 in the sandbox and 19 archived. Its members must also be corporate members of the LF; the members page listed four Premier members (Huawei, IBM, NVIDIA and ZTE), 16 General members, among them AMD, Cloudera, Databricks, Neo4j and SpiralDB, and 45 Associates, mostly universities and non-profits.

Like the CNCF, it doesn’t write the code. Each project keeps its own technical charter and steering committee and sits in its own series of LF Projects, LLC: Delta Lake’s site reads “Delta Lake, a series of LF Projects, LLC”. Most of its projects are AI tools, such as ONNX, Feast, Flyte and Docling, but it also hosts data infrastructure: the Delta Lake table format, which Databricks moved into it in May 2024 after four and a half years as a stand-alone LF project, Databricks’ open-source Unity Catalog, the LakeSoul lakehouse framework, the Vortex columnar file format, the Milvus vector database, the JanusGraph graph database and the OpenLineage lineage standard.

For someone choosing a technology, the badge says the name belongs to the foundation: the charter requires every project trademark to be transferred to LF Projects, LLC, and an archived project’s trademarks stay there. The code is under an OSI-approved licence, though the foundation doesn’t require one licence for all projects, as the CNCF requires Apache 2.0. The stage says less than a CNCF stage: its criteria are counts of contributing organisations, GitHub stars and OpenSSF badges, not interviews with adopters, and the foundation’s own lists don’t always agree on a project’s stage. Nor is a badge a verdict that a project is the best of its kind: it hosts both Delta Lake and LakeSoul, two competing table formats.

Its projects in this catalog

In depth

Sandbox to graduation, and the archive#

The criteria come from the project lifecycle document in effect since June 1, 2023. Counts are from the foundation’s projects page on October 28, 2026; two more entries, DocLang and Egeria Conformance, carry no stage.

Sandbox

18 projects

A project needs a sponsor that is already a member, an OSI-approved licence, documented open governance in a technical charter, and a contribution agreement that transfers its assets to the Linux Foundation. Before the vote it must move its code into its own GitHub organisation with @thelinuxfoundation as co-owner, turn on two-factor authentication and the DCO app, earn the OpenSSF Best Practices passing badge and add files such as GOVERNANCE.md, RELEASE.md and SECURITY.md. A majority vote of the TAC admits it. LakeSoul has been in the sandbox since May 2023 and Unity Catalog since June 2024.

Incubation

23 projects

On top of the sandbox requirements, at least three organisations must be actively contributing, a technical steering committee with a named chair must run the project, and it needs 500 GitHub stars and the OpenSSF Silver badge. A majority TAC vote promotes it, and it may then call itself an “LF AI & Data Incubation Project” and get design and marketing help. Projects can stay a long time: Feast has been incubating since September 2020 and JanusGraph since December 2020. Delta Lake joined at this stage in May 2024; Vortex was promoted in August 2025.

Graduated

12 projects

A graduating project needs code contributions from at least five organisations, 1,000 stars, the OpenSSF Gold badge, a substantial flow of commits over the past 12 months and at least one collaboration with another LF AI & Data project; the TAC has the final judgment on activity levels. Both the TAC and the Governing Board must vote for it, and it then gets a voting seat on the TAC. ONNX was accepted straight in as a graduated project in October 2019; Milvus graduated in June 2021 and Bitol, a data-contract standard, in July 2026.

Emeritus

19 projects

The TAC reviews every project each year and may move one that has gone inactive, or that it no longer supports, to the emeritus stage. A proposal is presented at a TAC call, stays open for at least two weeks of discussion and needs two-thirds of both the TAC and the Governing Board. The foundation then lists the project as archived and stops supporting it, but its assets and trademarks remain with LF AI & Data and the Linux Foundation. Acumos AI, the first project, is archived, as are Horovod, which graduated in September 2020, and Amundsen.

How a project gets in and moves up#

A project proposes itself with a pull request to the lfai/proposing-projects repository, using the foundation’s proposal template, and a TAC member is appointed as its sponsor. The donor has to be a member: the foundation says that if a donating company isn’t one, it first has to join LF AI & Data and the Linux Foundation, which is why SpiralDB, Vortex’s donor, is a General member. A project turned down for a higher stage may apply again after six months. Changes to the lifecycle document itself need votes of both the TAC and the Governing Board.

The foundation’s records lag its votes. Vortex was donated as a sandbox project in June 2025 and promoted to incubation in August 2025, when the LF’s press release called it “a new Incubation-stage project”, but the landscape data, last refreshed on October 15, 2026, still marked it sandbox. The foundation’s “incubation levels” web page still lists lower bars than the 2023 lifecycle document, two contributing organisations and 300 stars for incubation, and the project table in the proposals repository ends in October 2024 with DeepSpeed, which has been a PyTorch Foundation project since May 2025 and no longer appears on the projects page.

Who decides and who pays#

The Governing Board manages the fund: it approves the budget, elects a Chair and a Treasurer, and approves new projects in addition to the TAC. Its voting members are one representative of each Premier member, the TAC chair, and General-member representatives elected each year, one for every eight General members up to three, with at least one. No group of related companies may have more than two representatives on the board. Decisions at a meeting need a simple majority with half the voting representatives present, and amending the charter needs two-thirds of the whole board and the LF’s approval. Day-to-day work is done by LF staff: when IBM contributed Docling, Data Prep Kit and BeeAI in April 2025, the LF’s Todd Moore was the foundation’s interim executive director.

The Technical Advisory Council coordinates the projects, recommends where money should go and maintains the lifecycle. Its voting members are one representative of each Premier member and one from each “TAC Project”; graduation brings the right to such a seat, so the four Premier companies vote alongside up to 12 graduated projects. The charter sets annual dues of $100,000 for Premier members, $5,000 to $25,000 for General members depending on headcount, and nothing for Associates, which must be non-profits, open-source projects or government bodies. LF membership is paid on top, and the LF takes a fee of 9% of the fund’s first $1 million of receipts each year and 6% above that.

Licence, trademarks and contributions#

The charter requires any trademark relating to the fund or a project to be transferred to and held by LF Projects, LLC under its trademark policy. A sandbox project signs a contribution agreement transferring its assets to the Linux Foundation, gives the LF co-ownership of its GitHub organisation and installs the DCO app on every repository, so contributors sign off their commits. Projects use the LF Projects code of conduct unless another is approved, and are encouraged to put SPDX licence identifiers in every source file.

The licence is set by each project, within the OSI-approved rule: Delta Lake, Unity Catalog, LakeSoul and Vortex are all under Apache 2.0. The company that contributed a project can still lead its development and build products on it, as Databricks does with Delta Lake and Unity Catalog and Spiral with Vortex, but the name and the GitHub organisation are no longer its own.

Where it came from2018–2026

The Linux Foundation launched LF Deep Learning on March 26, 2018, with Amdocs, AT&T, B.Yond, Baidu, Huawei, Nokia, Tech Mahindra, Tencent, Univa and ZTE as founding members. Its one project, Acumos AI, a platform for sharing AI models, came from code by AT&T and Tech Mahindra. Angel, Elastic Deep Learning, Horovod and Pyro followed within a year, and on May 22, 2019 the foundation renamed itself LF AI to cover machine learning and AI beyond deep learning.

ODPi had spent five years on standardising the Hadoop and big-data ecosystem, and had moved on to metadata and governance with Egeria. On October 26, 2020 LF AI and ODPi announced they would come together as the LF AI & Data Foundation, under a single Technical Advisory Council, saying AI and data were “inseparable and codependent”. LF AI then had 25 members and 20 projects and was adding about one a month; Egeria and OpenDS4All became hosted projects.

Since then the data side has grown: LakeSoul joined in 2023, Databricks moved Delta Lake and Unity Catalog in during 2024, and Vortex arrived in 2025. Most newcomers are still AI tools, such as IBM’s Docling, Data Prep Kit and BeeAI in April 2025, and projects also leave: DeepSpeed, accepted in October 2024, became a PyTorch Foundation project in May 2025, and Horovod, which graduated in 2020, is archived.

Why it’s built this way

One foundation for AI and data
LF AI and ODPi argued that AI and data depend on each other, and that one Technical Advisory Council and one membership would bring closer integration between projects and cost members less.
Graduated projects vote on the TAC
The charter gives the TAC the job of coordinating collaboration among projects, so mature projects sit alongside Premier members, and graduation itself requires a collaboration with another hosted project.
Projects keep their own governance
Each project has its own technical charter and steering committee in LF Projects, LLC; when ODPi merged in, its projects kept their existing open technical governance while the foundation held the trademarks.

Timeline

  1. 2018

    LF Deep Learning launched on March 26 with Acumos AI as its only project

  2. 2019

    Renamed LF AI Foundation in May; ONNX accepted as a graduated project in October

  3. 2020

    Merger with ODPi into LF AI & Data announced October 26; Egeria joins

  4. 2023

    Current project lifecycle takes effect June 1; LakeSoul joins the sandbox in May

  5. 2024

    Databricks contributes Delta Lake (May) and Unity Catalog (June)

  6. 2025

    IBM contributes Docling, Data Prep Kit and BeeAI in April; Vortex promoted to incubation in August

  7. 2026

    Bitol graduates in July; 53 active and 19 archived projects in October

Sources · facts checked 2026-10-28

Improve this page on GitHub