Open-source AWS Step Functions alternatives

Conductor is an event-driven workflow engine that makes microservice and AI agent execution durable. It treats choices made at runtime, including loops, branches, parallel fan-out, tool calls, approval steps, retries and cancellation, as durable, inspectable executions. The project began at Netflix and is now maintained by Orkes and the community. It is written in Java with a JavaScript UI and released under the Apache-2.0 license.

Every step is persisted, so a workflow can continue after a crash, a restart or a network outage, and retries and timeouts are configurable. Orchestration is kept as a versioned graph while workers and built-in tasks perform the business logic. For AI use, it provides native LLM tasks, MCP tool calling, human approval steps and vector workflows for retrieval-augmented generation. A quickstart needs Node.js and Java 21 or later, serves a UI on port 8080, and a Docker image is also available.

Kestra is an orchestration platform, open source, for data pipelines, AI workflows and infrastructure automation. It brings scheduled and event-triggered automation together under one declarative interface that does not depend on a programming language, applying infrastructure-as-code practices to pipelines so that reliable workflows can be defined in a few lines of YAML.

Workflows can be built in the UI, written by a built-in AI Copilot, or generated from coding agents such as Claude Code and Cursor using agent skills, and everything can be kept as code with Git integration even when it was authored visually. A large plugin ecosystem connects it to external systems, and key concepts cover tasks, triggers and flows. The project is written in Java, with topics covering orchestration, high availability, pipeline-as-code and DevOps, and the 2.0 release adds new capabilities.

Kestra is Apache-2.0 licensed and can be self-hosted, with a quick start that gets a first workflow running in minutes. It targets data engineers, platform teams and automation developers who want an alternative to schedulers such as Airflow with a more declarative style.

Temporal is a durable execution platform that helps developers create scalable applications while keeping both productivity and reliability. The Temporal server runs units of application logic called Workflows, handling intermittent failures and retrying failed operations automatically. It started as a fork of Uber's Cadence and is developed by Temporal Technologies, a company founded by Cadence's creators. The server is written in Go and released under the MIT license.

This repository contains the source of the server only. To implement Workflows, Activities and Workers you use one of the supported language SDKs. The getting started guide shows how to download and run a pre-built image with its dependencies locally, try samples written for Go or Java, interact with the server through the Temporal CLI, and watch workflows in the Web UI at port 8233. Repository topics cover workflow engines, distributed cron, microservice orchestration and durable execution.

Argo Workflows is a container-native workflow engine for running parallel jobs on Kubernetes. It is implemented as a Kubernetes Custom Resource Definition, so workflows are defined as Kubernetes resources in which each step is a container.

Multi-step workflows can be modeled as a sequence of tasks or as a directed acyclic graph that captures dependencies between tasks. Typical uses include machine learning pipelines, data and batch processing, infrastructure automation and CI/CD. Client libraries exist for Java, Go, Python through the Hera SDK, and TypeScript.

The project is a Cloud Native Computing Foundation graduated project written in Go and licensed under Apache 2.0. It is cloud agnostic and runs on any Kubernetes cluster, and the ecosystem includes tools such as Argo Events, Kubeflow Pipelines and Katib that use or build on it. Interactive training material and a demo environment are linked for trying it out.

Inngest is a workflow orchestration platform built around durable functions. These functions replace separate queues, state management and scheduling so developers can write reliable step functions and background jobs without managing the underlying infrastructure.

An Inngest function has three parts. Triggers are events, cron schedules or webhook events, flow control configures how runs are queued and executed through concurrency, throttling, debouncing, rate limiting and prioritization, and steps are the building blocks that can retry on error and run for months. You write functions with a language SDK, test them with the Inngest Dev Server, deploy them to your own infrastructure and sync them with the Inngest Platform or a self-hosted server, which then calls them securely over HTTPS when events arrive.

Inngest is written in Go. Its quick starts cover Next.js, Node.js and Python, and the documentation includes project architecture and self-hosting guides. The repository lists its licence as other, so check the terms before running it yourself.

Dagu is a local-first workflow orchestrator for operations teams and internal automation. It is distributed as one executable that includes a web interface, requires neither a separate database nor a message queue, and works on Linux, macOS and Windows. It is aimed at teams whose main work is not orchestration, who just need dependable scheduling for existing scripts.

Workflows are directed acyclic graphs written in declarative YAML, keeping workflow definitions separate from business logic. Steps can invoke shell commands, containers, Kubernetes Jobs or commands on remote hosts over SSH, and Dagu Actions add further step types. It adds scheduling, retries, human approval tasks and run history, turning ad hoc scripts and runbooks into production workflows. Topics also mention MCP, AI agents and durable execution.

Dagu is written in Go and licensed under GPL-3.0. The project positions it as an alternative to Airflow, cron and job schedulers for people who want something lighter. A live demo, documentation, CLI and API references are available, and it can run on modest hardware.