BELLEVUE, Wash., Aug. 04, 2026 (GLOBE NEWSWIRE) — Union.ai today announced the general availability of Flyte 2, a ground-up rebuild of its open-source platform for AI and ML engineers. Flyte 2 provides durable, infrastructure-aware execution for engineering teams, enabling AI workloads to recover automatically from failures and scale to production.

AI workloads are increasingly long-running, compute-intensive, and non-deterministic, making them prone to infrastructure-caused failures. While traditional orchestrators can replay code logic after a failure, they cannot fix underlying issues such as out-of-memory (OOM), node interruptions, and container preemption. Flyte 2 introduces the first durable AI runtime in open source, capable of understanding and adapting the infrastructure and compute beneath the code. As a result, workflow failures that previously required manual intervention from engineers can now be automatically solved at runtime. This new level of durability makes it dramatically easier for engineers to scale AI workloads to production.

“AI teams need to iterate fast and scale workloads massively, but infrastructure reliability has been a major obstacle,” said Ketan Umare, co-founder and CEO of Union.ai. “We developed Flyte 2 to use infrastructure-as-context, so engineers can automatically recover workloads from failures, even if they’re caused by compute resources.”

Flyte 2 delivers on the durable AI runtime vision through three capabilities designed to keep AI workloads running reliably in production:

  • Durable execution, even for infra-caused failures: Flyte 2 provides a single durable execution model for logic and compute failures across training, inference, and agentic workflows. Self-healing workflows can diagnose failures, adjust resource requests on retry, and resume from saved progress, reducing the cost of manual recovery or restart.
  • Dynamic, agent-native execution: In pure Python, engineers can orchestrate agentic workflows that branch, loop, call tools, and adapt at runtime. Flyte 2 embeds durability into these traditionally brittle workflows.
  • Real-time and batch inference: Flyte 2 handles inference natively, so teams do not need to stitch a separate serving stack. The same runtime manages training, orchestration, and serving.

Together, these capabilities give teams a durable execution layer that adapts as their AI systems evolve, enabling faster iteration and easier scaling in production.

Flyte 2’s 1.3 million pre-GA downloads build on Flyte 1’s open-source foundation, whose SDK has surpassed 16 million downloads and is used by more than 4,000 organizations, including frontier AI labs and Fortune 500 companies. Union.ai’s partnerships with industry leaders, such as NVIDIA, Amazon, Google, and Nebius, further extend its ecosystem across the modern AI development stack.

“Union.ai and Flyte are critical for our AI development,” said Luca Del Pero, Director of Engineering, Head of AD/ADAS Cloud and Data Engineering, Woven by Toyota. “They allow us to build complex AI workflows and run them at scale without worrying about orchestration.”

As of today, Flyte 2 can be deployed to a remote cluster for production. Flyte 1 will continue to receive security-related updates through the end of 2026.

Union.ai has published a migration guide that includes migration instructions, automations, and live office hours. For a limited time, teams migrating to Flyte 2 from legacy orchestrators can also request a free consultation with Union.ai’s engineering team.

Learn more and try Flyte 2 at flyte.org.

About Union.ai
Union.ai provides the durable, infrastructure-aware runtime for scaling AI workloads to production. It is the only AI runtime that runs entirely in the customer’s cloud(s). Union.ai is the enterprise platform for Flyte, the first open-source AI runtime. Learn more at www.union.ai.

Media Contact:
UnionAI@merrittgrp.com


Primary Logo

About The Author