Flyte
Open-source orchestration engine for ML, data, and agentic AI workflows.
About
Flyte is an open-source workflow orchestration engine built for machine learning and data workloads, created at Lyft and now commercially backed by Union.ai. It treats workflows as typed, versioned, reproducible pipelines with caching, and it has been the ML orchestrator of choice for companies like Amazon, LinkedIn, and Stripe, with tens of millions of downloads.
In September 2025 Union.ai unveiled Flyte 2.0, a rebuilt engine for dynamic, long-running agentic workloads: pure-Python authoring instead of static DAG DSLs, runtime branching and loops, live debugging with interactive reruns, and resource-aware scheduling. Union.ai closed a $38.1M Series A led by NEA in February 2026 and reported 3x revenue growth in 2025. Flyte 2.0 was in private beta via Union's platform at launch — self-hosted open-source users should expect a maturing, still-evolving 2.x line.
What it can do
- Typed, versioned, reproducible ML workflows in pure Python
- Dynamic branching, loops, and runtime resource allocation
- Live debugging with interactive reruns and full observability
- Managed spot/preemptible handling and caching
- Flyte 2.0 engine for long-running agentic workloads
Setup
Self-host via flyteorg/flyte Helm charts on Kubernetes, or use Union's managed platform; workflows authored in Python with flytekit.
Prerequisites
Python 3.9+, a Kubernetes cluster for self-hosting (or a Union account).
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 77; Trending score 82/100 from researcher assessment.
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.