Back to catalog
Kestra
Apache-2.0Kestra replaces cloud orchestration like Prefect Cloud with a self-hosted, visual platform running declarative YAML workflows inside your private network.
Open-source alternative to — compare it against:
Overview
French engineer Ludovic Dehon built Kestra after managing data operations at retailer Leroy Merlin, where existing workflow engines proved too code-heavy for cross-functional teams. He designed the platform around declarative configurations and an embedded browser editor, allowing engineers and business operators to orchestrate tasks without custom Python code. Today, that foundation gives organizations infrastructure sovereignty by running event-driven workflows entirely inside private networks.
Why Kestra instead of Prefect Cloud
Prefect Cloud relies on a hosted service where pipeline graphs, scheduling metadata, and execution logs stream to vendor servers. While its hybrid agents run locally, the control plane and access permissions remain bound to external infrastructure. Kestra removes that dependency by deploying its complete stack—scheduler, web interface, and execution workers—directly into private Kubernetes clusters, Docker hosts, or air-gapped servers. Sensitive database records and execution parameters never leave internal perimeters.
Choosing Kestra shifts responsibility from subscription fees to infrastructure maintenance. Teams avoid per-user charges, but must manage database persistence and worker resources. Migrating from Prefect Cloud means replacing Python flow decorators with declarative YAML definitions. Teams lose dynamic Python code execution inside DAG definitions, but gain an operational console where non-programmers can trigger and monitor pipelines without installing terminal tools.
What it does well
- Visual workflow graphs display real-time execution states in the browser, making pipeline dependencies clear without inspecting source code.
- Declarative definitions live in a web editor with autocompletion, enabling operators to adjust parameters without Python tooling.
- Interactive execution forms let staff launch parameterized jobs on demand through clean web inputs rather than command lines.
- Event-driven triggers respond automatically to incoming webhooks and message queues without custom listener scripts.
- Centralized logs and task outputs surface directly in the interface, letting non-developers inspect run failures and download artifacts.
- Language-agnostic tasks run SQL queries, shell commands, and containerized tools without requiring code in a specific language.