Prefect is a Python-native workflow orchestration platform — open source at the core, managed via Prefect Cloud — that runs data, ML and AI agent workflows with durable execution and built-in recovery.
Data teams traditionally replace cron jobs with brittle scripts or wrestle with DAG frameworks that force workflows into rigid declarations. Prefect takes a different route: you write workflows as ordinary Python functions decorated with @task and @flow. Because there is no DAG to declare up front, your flows can branch, loop and generate tasks at runtime like normal code — which also means any coding agent that writes Python can produce a working Prefect flow without learning a DSL.
Durable execution is built into the engine rather than bolted on: scheduling, retries, caching and crash recovery are platform capabilities, not code you maintain. In production, Prefect Cloud provides a managed control plane with scheduling, observability, alerting and governance, while Prefect Serverless runs flows on managed compute so teams start with zero infrastructure. Teams with stricter requirements use the hybrid model — lightweight workers inside their own environment execute code where the data lives, while the cloud handles scheduling and visibility, an arrangement that clears most security reviews. Built-in dbt integration, multi-agent coordination workflows and FastMCP (Prefect’s open-source MCP server framework) round out an increasingly agent-oriented stack.
The Prefect framework is open source (Apache 2.0) and free to self-host. Prefect Cloud starts with a free Hobby tier (2 users, 5 deployments, 500 minutes of Serverless compute per month). The Starter plan costs $100/month (3 users, 20 deployments, 75 hours of Serverless, bring-your-own-compute), Team is $100/user/month with service accounts and audit logs, and Enterprise is custom with SSO, RBAC and IP allowlisting. Pricing is per developer, not per workflow or run.
Data and ML engineers who want orchestration in the Python they already write, and teams orchestrating AI agent workflows in production. Teams committed to DAG-centric stacks or needing purely serverless function platforms may prefer alternatives.