Dagster

Overview

Dagster orchestrates data assets rather than tasks, and that choice shapes everything else. You declare the tables, datasets, models and reports that should exist, as Python functions, and Dagster works out what to build and in what order. Lineage and observability follow from the declarations rather than being bolted on.

The practical difference from a task-first orchestrator is what you see when something breaks. The graph is of data, not steps, so you can ask which asset is stale and what depends on it.

Asset checks are part of the model: a check runs against an asset to validate nulls, schema or other properties, and a blocking check stops downstream materialisation when it fails, which puts data quality inside orchestration rather than beside it.

Automation is declarative too. Rather than only cron, you attach conditions to assets and let Dagster decide when to materialise them, with schedules, sensors and asset sensors also available.

Testability was an explicit design goal, and the documentation carries dedicated guides for testing assets.

Dagster Open Source is Apache 2.0. Dagster+ adds branch deployments, hybrid execution, roles, cost insights and observability, either serverless or running agents in your own infrastructure.

One thing anyone comparing should know: Prefect announced in July 2026 that it is acquiring Dagster Labs. Dagster keeps its name, licence and commercial service, and the company says the open-source project will continue to be actively developed, but the two products now share a corporate owner. Its founder has stepped down.

Key features and capabilities

The same headings are used for every orchestration & scheduling entry, so two tools can be read side by side.

How pipelines are written

• Assets declared as Python functions with a decorator; code is the source of truth
• Multi-asset definitions for one computation producing several assets
• Ops and jobs remain available for task-style work
• Dedicated guides for testing assets and writing asset checks

Scheduling and triggers

• Cron schedules over a selection of assets
• Declarative automation: conditions on assets decide when they rebuild
• Sensors and asset sensors for event-driven runs
• Partitions and backfills for time or category slices

Execution and scaling

• Self-hosted locally, as a service, on Kubernetes, Docker Compose, AWS or Google Cloud
• Dagster+ serverless runs your code in their environment; hybrid runs it in yours
• Hybrid agents: Kubernetes, Amazon ECS, Docker, Azure or local

Monitoring and recovery

• Web interface with run history, logs and debugging
• Asset health and freshness surfaced alongside runs
• Dagster+ adds cost insights and branch deployments

Data awareness

• The asset graph is the core model; assets know their dependencies, ops do not
• Asset checks validate an asset and can block downstream materialisation
• Freshness policies under the observe section
• OpenLineage among the integration libraries

Integrations

• Around 90 integration libraries, official and community
• dbt, Spark, Snowflake, Databricks, BigQuery, Iceberg, Delta Lake and Polars
• Airbyte, Fivetran, dlt, Sling and Meltano for ingestion; Looker, Power BI and Tableau for BI
• Great Expectations, Soda, DataHub, Atlan and OpenMetadata for quality and cataloguing

How it runs

• Open source self-hosted; Dagster+ serverless or hybrid
• Kubernetes, Docker Compose, AWS and Google Cloud documented
• Apache 2.0 for the open-source project; Dagster+ is commercial

Pricing

Usage-basedSolo $10 a month plus credits

Open source is free under Apache 2.0. Dagster+ is priced in credits, where one asset materialisation or op execution is one credit. Solo is $10 a month plus $0.040 a credit for a single user; Starter is $100 a month plus $0.035 a credit for up to three users; serverless compute is $0.010 a minute. Pro and Enterprise are quoted, adding unlimited deployments, cost tracking, single sign-on and audit logs. Both paid entry tiers offer a 30-day trial.

Vendor pricing page →

Demos and videos

• Dagster+ demo: asset health monitoring and team workflows Product demo · YouTube
• Dagster channel Channel or playlist · YouTube

About Dagster Labs (being acquired by Prefect)

Dagster Labs was founded in 2018 by Nick Schrock, previously a co-creator of GraphQL at Facebook. The project is Apache 2.0 but vendor-led rather than foundation-governed. In July 2026 Prefect announced it was acquiring the company: around 40 staff move across, the founder is stepping down, and the combined company operates under the Prefect name, though Dagster keeps its brand, contracts and support. The company's about page now redirects to Prefect's. Head office and funding history are not published on its current pages.

Founded 2018 · dagster.io