AI Potluck
Contribute
Back to Gap Map Infrastructure / ML orchestration

Flyte

Union.ai
open source / Overall score: 3.0

Kubernetes-native orchestrator for machine-learning and data workflows, written as typed Python tasks. Flyte's control plane, Propeller, walks the workflow graph and emits one pod per task against the resources that task declared, caching results and passing typed artifacts between steps. Plugin task types cover PyTorch, Spark, Ray and Dask training jobs alongside plain containers.

Union.ai, founded by Flyte's creators, sells a managed platform beside it and hosts the project's documentation.

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public(the control plane and the SDKs both build from this repo)
core features withheld
no — a 2,504-entry recursive tree carries a root LICENSE plus flyteplugins/LICENSE and flytestdlib/LICENSE, and no path matching enterprise, ee, commercial or proprietary

Flyte's orchestrator is published under Apache-2.0, and Union.ai's managed platform sits beside it as a separate commercial product rather than a gate on the core. A full scan of the repository tree finds a governing Apache-2.0 license plus two component licenses under flyteplugins and flytestdlib, and no enterprise, ee or commercial path.

  • https://api.github.com/repos/flyteorg/flyte/git/trees/main?recursive=1 recorded 2026-09-15

    Full untruncated recursive tree of the default branch, 2,504 entries. Licenses are the root LICENSE plus flyteplugins/LICENSE and flytestdlib/LICENSE. No path matches enterprise, ee, commercial or proprietary. A tree lists paths; it is cited for that and for nothing about the vendor's offerings.

  • https://github.com/flyteorg/flyte recorded 2026-09-15

    Repository page for flyteorg/flyte: Apache-2.0 license, public and unarchived, described as 'Dynamic, resilient AI orchestration'. Establishes the license and that the source is public.

  • https://www.union.ai/ recorded 2026-09-16

    Union.ai's homepage, presenting its own runtime as a separate commercial product deployed into the customer's cloud rather than as a fuller edition of Flyte. Establishes that the commercial offering sits beside the open orchestrator.

Adoption

3 high confidence
3.0

Adoption is measured as PyPI downloads summed across the flyte and flytekit packages in the trailing month. A user who installs both is counted twice, but each package individually already reads the same as the sum, so the double count does not change the reading.

Capability

3 medium confidence
3.0

Flyte's own control plane decides when each task runs and which cluster and project-domain it goes to and emits the pod, then the Kubernetes scheduler picks the machine, the same shape of decision as Metaflow. The two sit together because the evidence does not separate them further.

  • https://github.com/flyteorg/flyte recorded 2026-09-15

    Repository page and README for flyteorg/flyte describing the control plane, typed tasks and the plugin task types.

Verified 2026-09-15