Moving from Python-heavy orchestrators to Kestra’s YAML pipelines

The shift to declarative workflows

I recommend Kestra for teams that want to move away from the heavy Python requirements of Airflow or Prefect. While Airflow relies on Python-based Directed Acyclic Graphs and Prefect focuses on Python-first dynamic workflows, Kestra uses declarative YAML configuration files to define all flows. This approach separates orchestration logic from business logic. I find that Kestra provides a different path for teams that find the Python requirements of Airflow or Prefect too restrictive when they need to run tasks in languages like R, Go, or SQL. You can build flows using the YAML editor, the No-Code editor, or the AI Copilot. These three methods all edit the same flow and stay in sync. Kestra was founded in 2021 by Emmanuel Darras and Ludovic Dehon. The platform is JVM-based and can be deployed on Kubernetes, Cloud Compute, Docker, or on-premises. Workers communicate with the control plane over gRPC, which helps the platform scale. The platform has more than 26,000 stars on GitHub. Kestra 2.0 includes a new distributed execution engine built for reliability at scale. The platform includes over 1,500 integrations and tasks. Note that Airflow 3.3.0 shipped in July 2026 with a Language Task SDK for Java and Go, which may influence some teams to stay.

Scale and enterprise adoption

Kestra runs on a message-driven architecture that uses Apache Kafka and ElasticSearch to manage workloads. This design allows the platform to handle millions of workflows. In 2025, Kestra executed over 2 billion workflows, an increase of 20 times over the previous year. Kestra manages workflows through an executor implemented with Kafka Streams. The company raised 36 million dollars in total funding by March 2026. This includes a 25 million dollar Series A led by RTP Global and an earlier 8 million dollar round. Since its seed round, Kestra has seen enterprise revenue grow 25 times. The team consists of 20 people spread all over Europe. Large organizations like Apple, Toyota, and JPMorgan Chase use Kestra for their production pipelines. Apple uses it for machine learning, whereas Toyota unifies data and AI pipelines. The Enterprise edition includes RBAC, audit logs, and multi-tenancy. Kestra also handles millions of orchestration events per month for clients like Leroy Merlin and Decathlon. The company is expanding into North America and Europe to support its growth. Kestra Cloud is a fully managed SaaS experience with usage-based pricing.

Triggers and installation

You can start Kestra locally in minutes using Docker. Run the docker-compose command to pull the necessary images and start the server. Once the server runs, access the interface at localhost:8080. Kestra uses different trigger types to start flows automatically. You can use a Schedule trigger with cron expressions, a Webhook trigger for inbound HTTP requests, or a Polling trigger to detect changes in external systems. The platform also includes a Flow trigger to run workflows when upstream flows complete. Kestra also supports Realtime triggers with millisecond latency and an MCP Tool trigger so AI agents can call flows. I find the automation of file handling between tasks to be a clear advantage. Kestra automatically downloads files from a trigger and passes the URI to the next task. The subflow function allows you to return outputs inline. To stop a trigger after a specific state like SUCCESS, use the stopAfter property. You can disable a trigger via the UI. Kestra 2.0 allows for draft revisions to stage changes. I find that the YAML syntax can be difficult because of indentation requirements. Do you think the indentation challenges of YAML will slow down your team?