AI Potluck
Back to Gap Map Infrastructure / ML orchestration

Union.ai

Union.ai
closed / Overall score: n/a

Commercial AI runtime built on Flyte, deployed into the customer's own cloud rather than hosted by the vendor. It runs the same typed Python workflows Flyte does, adds managed control-plane services around them, and keeps data and models inside the customer's infrastructure.

The commercial product beside Flyte, from the same organization; both are on this map, as ray and anyscale-fine-tuning already are. Verified 2026-09-15 via the Union.ai homepage and pricing page.

Openness

1 high confidence
1.0
license
Proprietary
service
proprietary(Union.ai commercial platform, deployed into the customer's cloud)
source
closed(no source is published for the Union control plane

Closed on the dimension's own question - whether the source of the product you run is published - rather than on its managed-service example. Union is self-hostable in the sense that it deploys into the customer's cloud, but no source repository exists for the control plane; the open flyteorg/flyte repository is Flyte, scored separately. Deployment location and source availability are different questions, and this dimension asks the second.

  • https://www.union.ai/ recorded 2026-09-15

    Union.ai's own homepage: 'The durable AI runtime that runs in your cloud, not ours', with zero-trust deployment into the customer's infrastructure. Establishes the deployment model and that no implementation is published.

  • https://www.union.ai/pricing recorded 2026-09-15

    Union.ai's pricing page, establishing that the product is sold commercially rather than published.

Adoption

not assessed

No usage figure is published for this service specifically, and it publishes no countable artifact - no package, repository or registry entry - so no level is assigned rather than one being inferred from the parent platform.

  • https://www.union.ai/ recorded 2026-09-15

    Union.ai's homepage, read for a usage disclosure; it names customers but publishes no count of workflows, runs or customers.

Capability

3 medium confidence
3.0

Banded at flyte, whose engine it runs. That is the honest placement rather than an independent judgement: under this dimension the two do the same thing, and the commercial control plane around it does not change which component chooses the machine. The Kubernetes scheduler still does.

Verified 2026-09-15