Skip to content
SPECIFICTION
Get listed

Flyte

flyte.org

From $950/mo Free plan Open source WebLinuxAPISelf-hosted
Visit flyte.org
Spec score 9.1 #1 of 29 in MLOps Platforms

What it is

Flyte 2 is a durable runtime for AI workloads involving training, serving and agents. Users write workflows as plain Python async functions and use a TaskEnvironment to declare images, resources, secrets and retry policies. Flyte resumes from failed tasks and can recover from out-of-memory kills, preempted nodes and disappearing GPUs. Branches, loops and runtime decisions support dynamic workflows and agents. Native serving integrations include Streamlit, FastAPI, vLLM, SGLang and Ollama, with replicas able to scale to zero when idle. Listed integrations also cover Apache Spark, Ray, PyTorch Elastic, BigQuery, Snowflake, Weights & Biases, OpenAI Agents SDK, LangGraph, CrewAI, Pydantic AI, MLflow, Dask and Databricks. It runs locally with a devbox or on a customer cluster. Union.ai says customer-cloud data planes keep workflow inputs, outputs, logs, code and reports outside its control plane. Flyte is available as free Apache 2.0 open-source software.

Spec sheet

What it does
Flyte is an open-source runtime for orchestrating AI workloads such as training, serving and agent workflows, with automatic recovery from failures.Read 15 Oct 2026 · flyte.org
Purpose
Flyte 2 is a durable AI runtime for workloads involving training, serving, and agents.Read 8 Oct 2026 · flyte.org
Deployment
Flyte 2 can run locally with a devbox or on a user's cluster; the site provides `pip install flyte` and `flyte start devbox` setup commands.Read 8 Oct 2026 · flyte.org
Integrations
The Flyte site lists integrations including Apache Spark, Ray, PyTorch Elastic, BigQuery, Snowflake, Weights & Biases, OpenAI Agents SDK, LangGraph, CrewAI, Pydantic AI, MLflow, Dask, and Databricks.Read 8 Oct 2026 · flyte.org
Security
Union.ai states that its data plane runs inside the customer's cloud and that workflow inputs, outputs, logs, code, and reports do not pass through its control plane.Read 8 Oct 2026 · union.ai
Support
The Team plan lists one business day Sev 1 support, while Enterprise lists custom white-glove support.Read 8 Oct 2026 · union.ai
License
The Flyte repositories and SDKs are published under the Apache License 2.0.Read 15 Oct 2026 · github.com
Maker
Union.ai describes itself as the enterprise Flyte platform and identifies Seattle, Washington as its location in its LinkedIn company profile.Read 8 Oct 2026 · union.ai
Autoscaling
Long-running apps can scale replicas up or down automatically and scale to zero when idle.Read 15 Oct 2026 · flyte.org
Batch inference
Flyte 2 includes a DynamicBatcher for batching inference requests with configurable batch size, queue size and timeout.Read 15 Oct 2026 · flyte.org
Caching and reproducibility
Executions are automatically logged and versioned, and identical inputs can reuse cached results.Read 15 Oct 2026 · flyte.org
Certifications
Union.ai's security page lists SOC 2 Type II, SOC 2 Type I, and HIPAA.Read 8 Oct 2026 · union.ai
CLI
Flytectl is a Go-based cross-platform command-line client for FlyteAdmin, with Homebrew and shell-script installation instructions.Read 15 Oct 2026 · github.com
Company facts
LinkedIn lists Union.ai as privately held, founded in 2021, headquartered in Seattle and employing 11–50 people.Read 15 Oct 2026 · linkedin.com
Company history
Union.ai's company page says Flyte was open-sourced in 2020 and Union.ai was founded with $10 million in seed funding in 2021.Read 15 Oct 2026 · union.ai
Deployment architecture
Legacy deployment documentation describes sandbox, single-cluster and multi-cluster Kubernetes deployment paths, with production use requiring a real object store.Read 15 Oct 2026 · docs-flyte-legacy.union.ai
Developer requirements
Legacy development documentation supports Python 3.8 or newer and installs Flytekit with pip.Read 15 Oct 2026 · docs-legacy.flyte.org
Documentation depth
The current documentation includes version selectors for Flyte 2 and Flyte 1 plus guides covering tasks, apps, agents, integrations and CI/CD deployment.Read 15 Oct 2026 · union.ai
Durable execution
Flyte 2 can recover from infrastructure failures including OOM kills, preempted nodes and disappearing GPUs.Read 15 Oct 2026 · flyte.org
Dynamic execution
Flyte supports branching, loops, and runtime decisions for dynamic workflows and agents.Read 8 Oct 2026 · flyte.org
Dynamic workflows
Workflows use ordinary Python control flow for branching, loops and runtime decisions rather than a separate DSL.Read 15 Oct 2026 · flyte.org
Extensibility
Flyte integrations are implemented through a plugin system and the project invites users to build and submit their own integrations.Read 15 Oct 2026 · union.ai
Failure recovery
Flyte can resume from a failed task and automatically recover from infrastructure failures such as OOM kills, preempted nodes, or disappearing GPUs.Read 8 Oct 2026 · flyte.org
Independent praise: Kubernetes fit
Coding Protocols praises Flyte's typed pipelines, Kubernetes execution and per-task resource control, recommending it for teams already deep on Kubernetes.Read 15 Oct 2026 · codingprotocols.com
Independent praise: typing and reproducibility
Reddit users repeatedly praise Flyte's enforced typing, caching, versioning and reproducibility for machine-learning workflows.Read 15 Oct 2026 · reddit.com
Limit: ecosystem gaps
A Reddit reviewer says some features were missing in Flyte 1 and that they still had to store artifacts in MLflow.Read 15 Oct 2026 · reddit.com
Limit: Kubernetes dependency
Independent comparison guidance positions Flyte primarily for teams already operating Kubernetes, implying a poor fit where that platform expertise is absent.Read 15 Oct 2026 · codingprotocols.com
Limit: operational complexity
Reddit reviewers report setup headaches involving permissions, configuration, firewalls and cloud deployment.Read 15 Oct 2026 · reddit.com
Logical recovery
Failed tasks can be retried and resumed with the same inputs instead of rerunning the whole workflow.Read 15 Oct 2026 · flyte.org
Migration status
Flyte 1 is scheduled to receive security patches through the end of 2026 while users migrate to Flyte 2.Read 15 Oct 2026 · flyte.org
News: AWS Marketplace
Union.ai announced AWS Marketplace self-service availability for its enterprise platform on September 30, 2026.Read 15 Oct 2026 · union.ai
News: local open source release
Union announced Flyte 2 local open-source execution on March 4, 2026, with hosted distributed execution offered through Union.ai production preview.Read 15 Oct 2026 · flyte.org
OAuth limitation
Flytectl OAuth requires a CA-certified SSL certificate or an external OAuth provider; self-signed certificates are rejected.Read 15 Oct 2026 · docs-legacy.flyte.org
Official integrations
The integration catalog lists Spark, Ray, PyTorch Elastic, BigQuery, Snowflake, MLflow, Weights & Biases, Dask, Databricks and many other plugins.Read 15 Oct 2026 · flyte.org
Origin
Flyte was built internally at Lyft in 2016, open-sourced in 2020, and joined the Linux Foundation AI & Data in 2022.Read 8 Oct 2026 · union.ai
Plan limits
The Team plan includes a 25,000 action-per-run fanout limit, 1,000 concurrent actions, and one cluster; Enterprise lists 50,000+ actions per run and 3+ clusters.Read 8 Oct 2026 · union.ai
Public API
Flyte services expose gRPC APIs and publish an OpenAPI schema accessible from the local API endpoint.Read 15 Oct 2026 · docs-legacy.flyte.org
Pure Python authoring
Flyte 2 tasks are asynchronous Python functions defined inside TaskEnvironments.Read 15 Oct 2026 · flyte.org
Rating on Capterra
4.8 out of 5 from 11 reviews, with the listing describing sales notes, competitive insights and revenue forecasting rather than Flyte OSS.Read 15 Oct 2026 · capterra.com
Rating on G2
4.7 out of 5 from 11 reviews, but the G2 listing describes Flyteai.com sales-coaching software rather than Flyte OSS.Read 15 Oct 2026 · g2.com
Rating on GetApp
4.8 out of 5 from 11 reviews, and the page likewise describes a sales and call-recording product named Flyte.Read 15 Oct 2026 · getapp.com
Real-time serving
Flyte 2 can deploy single-node real-time inference applications and exposes examples using FastAPI.Read 15 Oct 2026 · flyte.org
Sandboxing
Flyte 2 provides a sandbox for executing LLM-generated Python or shell code in ephemeral containers.Read 15 Oct 2026 · flyte.org
SDKs
Public repositories provide Python, Go, Rust, TypeScript and Mojo SDKs, while the Java/Scala SDK repository is archived.Read 15 Oct 2026 · github.com
Security practice
Security documentation describes running Flyte components as non-root users where possible and notes that some initialization containers still require root.Read 15 Oct 2026 · docs-legacy.flyte.org
Self-hosting
Flyte OSS documentation says users deploy and manage the platform on their own cloud infrastructure.Read 15 Oct 2026 · union.ai
Serving
Flyte's native app integrations include Streamlit, FastAPI, vLLM, SGLang, and Ollama, with support for scaling replicas to zero when idle.Read 8 Oct 2026 · flyte.org
Spot fallback
Interruptible tasks can use spot or preemptible instances with automatic fallback to on-demand capacity.Read 15 Oct 2026 · flyte.org
Type-checked I/O
The runtime supports language-native typed inputs and outputs including files, dataframes and dataclasses.Read 15 Oct 2026 · flyte.org
Workflow authoring
Flyte 2 lets users write workflows as plain Python async functions using a TaskEnvironment to declare images, resources, secrets, and retry policy.Read 8 Oct 2026 · flyte.org

Plans and prices

Flyte OSS Free Apache 2.0 open-source runtime · runs locally or on your cluster Read 8 Oct 2026 · flyte.org
Flyte OSS Free Free and open source; users deploy and manage it themselves Self-managed infrastructure · Flyte 2 documentation describes deployment on your own cloud infrastructure Read 15 Oct 2026 · union.ai
Team $950 a month $950/ mo + usage 30-day free trial · $950 monthly minimum credited to usage · 1,000 concurrent actions · 30-day data retention · 1 cluster Read 8 Oct 2026 · union.ai
Union.ai Team (enterprise Flyte platform) $950 a month Usage-based billing with a $950 monthly minimum credited toward usage; billing starts after the 30-day trial Metered durable actions · vCPU-hours · memory · GPU-hours · AWS Marketplace availability announced September 30, 2026 Read 15 Oct 2026 · union.ai
Enterprise Priced on request Custom Custom concurrent actions · custom retention · 3+ clusters · white-glove support Read 8 Oct 2026 · union.ai

Every Flyte plan and price →

Strong points and limits

Strong points

  • Python workflow authoring
  • Automatic failure recovery
  • Dynamic branches and loops
  • Broad AI integrations
  • Free Apache 2.0 runtime

Limits

  • Team plan costs $950/mo plus usage
  • Team has one cluster
  • Team limits fanout and concurrency
  • Enterprise pricing is custom

Who it suits

It suits teams running AI training, serving or agent workflows that need retries, dynamic execution and self-hosted deployment. The listed Team and Enterprise plans target organizations needing concurrency, clusters and support.

Flyte in depth

Read the full write-up

Flyte 2 is a Python runtime for coordinating durable AI workloads, including training, serving, and agents. It is best suited to teams comfortable running workflows on their own infrastructure or choosing a managed cluster plan. Its standout strength is recovery from task and infrastructure failures; the managed Team plan’s $950 monthly minimum is a substantial commitment.

Overview

Flyte combines workflow orchestration with the execution environment for AI workloads. Teams write workflows in Python and can deploy locally or on a cluster they control, with a hosted option from Union.ai for those who want a managed platform. The open-source runtime keeps the starting cost at zero, but operating a cluster remains part of the self-hosted path.

Flyte was built internally at Lyft in 2016, open-sourced in 2020, and joined the Linux Foundation AI & Data in 2022. Union.ai describes itself as the enterprise Flyte platform.

Key features

Workflows are plain Python async functions, with a TaskEnvironment for declaring images, resources, secrets, and retry policy. That keeps execution requirements close to workflow code, which is a natural fit for Python teams. Teams seeking a visual or non-code authoring approach should look elsewhere.

Flyte can resume from a failed task and recover automatically from infrastructure failures such as OOM kills, preempted nodes, or disappearing GPUs. That matters when rerunning work is costly or long-running; automatic retries and execution observability further support operations. Dynamic branching, loops, and runtime decisions let workflows adapt as they execute, a useful capability for agent workflows with changing next steps.

Native serving integrations include Streamlit, FastAPI, vLLM, SGLang, and Ollama, and replicas can scale to zero while idle. Its integration ecosystem also includes Apache Spark, Ray, PyTorch Elastic, BigQuery, Snowflake, Weights & Biases, OpenAI Agents SDK, LangGraph, CrewAI, Pydantic AI, MLflow, Dask, and Databricks. Event triggers broaden its orchestration options.

Flyte runs locally through a devbox or on a user's cluster. Union.ai says its data plane runs inside the customer's cloud, keeping workflow inputs, outputs, logs, code, and reports out of its control plane. Its security page lists SOC 2 Type II, SOC 2 Type I, and HIPAA.

Pricing

PlanPriceIncludesFlyte OSSfreeApache 2.0 open-source runtime; runs locally or on your cluster.Team$950 a month30-day free trial; $950 monthly minimum credited to usage; 1,000 concurrent actions; 30-day data retention; one cluster.EnterpriseCustom pricingCustom concurrent actions and retention, 3+ clusters, and white-glove support.Flyte OSS is the lowest-cost route for teams able to run their own infrastructure. Team adds a managed option, but its monthly minimum makes it a poor fit for small experiments or light workloads; its 1,000 concurrent actions and single cluster also set firm operating boundaries. Its 30-day trial gives teams a month to evaluate the plan. Enterprise suits larger deployments needing at least three clusters, customized capacity or retention, and tailored support. Team lists one-business-day Sev 1 support, while Enterprise offers custom white-glove support.

Platforms

Flyte supports API, Linux, web, and self-hosted use. The self-hosted deployment model is central to the OSS option: Flyte can run on a local devbox or a user's cluster. The managed plans are a better fit for teams that prefer Union.ai to provide the enterprise platform, while the data-plane design keeps customer workflow data inside the customer's cloud.

Who it's for

Flyte is a strong fit for Python-oriented AI teams running training, serving, or agent workflows that need retries, recovery, and dynamic execution. It is especially compelling when workflows must continue through interruptions or span an existing ecosystem of data, ML, and agent tools. It is less suitable for teams without cluster-operating capacity that also cannot justify the Team plan minimum, or for those who need workflow authoring outside Python.

Pros and cons

Pro: Task resumption and recovery from infrastructure failures reduce the need to restart long-running work after common cluster disruptions.

Pro: Dynamic branching, loops, and runtime decisions accommodate workflows and agents whose execution path changes as results arrive.

Pro: OSS runs locally or on a user's cluster, while a broad set of serving, data, and ML integrations supports varied workloads.

Con: Python async authoring narrows its appeal for teams that want a visual or non-Python workflow builder.

Con: Team starts at a $950 a month minimum, with one cluster and a 30-day retention period, which can constrain smaller or multi-cluster deployments.

Con: The free OSS route shifts deployment and cluster operation to the user.

Alternatives

Kestra is another free-plan workflow platform, with Linux, macOS, Windows, API, web, and self-hosted support; consider it when that platform mix is the better match. DataKitchen TestGen is worth considering when the need is specifically its open-source testing plan, which includes one user, one connection, and one project.

Apache NiFi offers a free Apache-licensed option with Linux, macOS, Windows, API, web, and self-hosted support, making it an alternative for teams prioritizing those deployment platforms. Apache Airflow is a free Apache-licensed choice with API, Linux, web, and self-hosted support. Apache Hop is free and runs on Linux, macOS, Windows, and self-hosted environments; its release 2.19.0 requires Java 21.

Metaflow is a free Apache-licensed option if teams want production scaling on infrastructure in their own cloud account or Kubernetes cluster. Dagster is another open-source orchestration framework with a paid Solo plan at $10 a month; choose it when its open-source core or Solo plan better fits your orchestration needs. Prefect has a free Hobby plan with two users, one workspace, five deployments, 500 minutes per month of Prefect Serverless, and seven-day run retention, making it a fit for smaller workloads within those caps.

For broader comparisons, see DataOps Platforms, MLOps Platforms, Workload Automation Software, and Workflow Orchestration Software.

Verdict

Choose Flyte if your team builds AI workloads in Python and needs durable, dynamic execution with a choice between local or self-managed deployment and a managed cluster. Its recovery capabilities and integration breadth are the strongest reasons to adopt it. Look elsewhere if you need non-Python authoring, want to avoid operating infrastructure without paying a substantial managed-plan minimum, or need more than one cluster on Team.

Verdict

Flyte provides workflow durability and dynamic execution across local or cluster deployments, with extensive integrations. Check plan limits, cluster needs and support requirements before choosing a hosted tier.

Where Specifiction places it

Alternatives to FlyteAll alternatives →