Flyte is an open-source orchestration platform for machine learning pipelines, model serving and AI agents, built to run reliably at scale on Kubernetes. Now at Flyte 2, it lets you define type-safe, distributed workflows in pure Python with async/await โ and it is a graduated project of the LF AI & Data Foundation. The project has earned about 7,700 stars on GitHub and is in active development in 2026.
Flyte treats pipelines as real software: tasks are typed Python functions, environments declare their own container images and resources, and the platform executes everything distributed across your cluster with caching, versioning and reproducibility built in. Flyte 2 streamlines this into a local-first experience โ you can run and debug the same code on your laptop with its devbox and TUI, then scale it out unchanged. It covers the full lifecycle in one system: data and ML pipelines, model training, real-time inference and model serving (including FastAPI apps), and agent runtimes with MCP support, with SDKs beyond Python for Rust, Go and TypeScript. For production backends today, the project’s commercial home, Union.ai, offers the enterprise-ready managed platform, while the open-source backend is developed in the open in this repository.
Flyte is free and open source under the Apache 2.0 license. An enterprise-ready, production-grade managed backend is available commercially from Union.ai for teams that want it hosted.
Flyte is for ML platform and data engineering teams that need reliable, large-scale orchestration of pipelines, models and agents on their own Kubernetes infrastructure.