Flyte OSS

Flyte is a free and open source platform that provides a full suite of powerful features for orchestrating AI workflows. Flyte empowers AI development teams to rapidly ship high-quality code to production by offering optimized performance, unparalleled resource efficiency, and a delightful workflow authoring experience. You deploy and manage Flyte yourself, on your own cloud infrastructure.

These are the Flyte 2.0 docs. To switch to version 1.0 or to the commercial product, Union.ai, use the selectors above.

Basics

Learn the basics of Flyte, covering all the core concepts around tasks and apps.

Flyte 2

Build AI workflows in pure Python with built-in durability, reproducibility, and recovery.

Quickstart

Install the SDK and run your first workflow locally in a few minutes.

Core concepts

The building blocks of every Flyte program: TaskEnvironments, tasks, runs, actions, and apps.

Run modes

Run the same task code locally, on a devbox, or on a remote cluster.

Tasks

Build durable, scalable, and reproducible batch workloads.

Configure tasks

Define TaskEnvironments for container images, resources, secrets, caching, retries, and more; use triggers for schedules.

Build tasks

Compose tasks with fanout, parallelism, error handling, traces, files, and DataFrames.

Run and deploy tasks

Use flyte run for iteration or flyte deploy to register a stable task version.

Apps

Create long-running services to host dashboards, APIs, and model endpoints.

Configure apps

Define AppEnvironments with ports, autoscaling, custom domains, and authentication.

Build apps

Build dashboards, REST APIs, and model endpoints with FastAPI, Streamlit, vLLM, and more.

Native app integrations

Use pre-built environments for popular frameworks like Streamlit, FastAPI, vLLM, and SGLang.

Serve and deploy apps

Use flyte serve for fast iteration or flyte deploy for production deployments.

Agents

Build durable, self-healing agents using tasks and apps as building blocks.

Build agents

Implement ReAct, Plan-and-Execute, and other agent patterns with full observability.

Agent framework integrations

Integrate with third-party agent frameworks like LangGraph, PydanticAI, and OpenAI Agents SDK.

Sandboxing

Safely execute LLM-generated code with workflow sandboxes or ephemeral containers.

Build an MCP

Serve Model Context Protocol servers for AI assistants to interact with, hosted on Flyte.

Advanced guides

Organize your codebase, optimize performance for production, and migrate from other workflow orchestrators.

Project patterns

Patterns for BYO images, monorepos with uv, CI/CD, and multi-team resource management.

Run scaling

Tune task overhead, batching, reusable containers, and fanout to scale your workflows.

Advanced project

An advanced guide for building an LLM reporting agent on Flyte.

Migration

Port a Flyte 1 codebase to Flyte 2, or map Airflow concepts to their Flyte 2 equivalents.