Announcing StepFlow: An Open Protocol and Runtime for GenAI Workflows

We're excited to announce the initial release of StepFlow, an open-source protocol and runtime designed to make building, executing, and scaling GenAI workflows simple, secure, and portable. Whether you're prototyping locally or deploying to production, StepFlow provides the foundation for reliable AI workflow execution.

Why StepFlow?

The rapid evolution of GenAI over the past couple of years has created an explosion of tools, models, and APIs. While this diversity has powered some serious innovation, it also creates challenges: How do you build workflows that can leverage different AI services? How do you ensure your local prototypes will scale in production? How do you maintain security and isolation when executing untrusted code? How do you parallelise your workflows to maximise resource utilization?

LangFlow - one of our other projects - meets (exceeds even!) most of these requirements with both a visual authoring experience and a built-in API server that turns every agent into an API endpoint. These endpoints can then be integrated into applications built on any framework or stack, supporting all major vector databases, LLMs, and a variety of other AI tools.

However, it became clear to us that a pure server implementation for running “flows” was necessary to really scale and provide isolation for security and resource consumption. StepFlow was designed to fill this gap, not only for LangFlow but every other workflow system, and address these challenges with a simple yet powerful approach: define your workflows in YAML or JSON, and let the runtime handle the complexity of execution, scaling, and security.

Key Features in This First Release

Reliable Workflow Execution for Everyone

StepFlow focuses on three core users: workflow creators, workflow frameworks, and workflow platforms. Workflow creators can use more components, run in more places, and scale more reliably. Workflow frameworks can leverage all the existing components without the need to deal with scale and durability. Workflow platforms can run workflows from any supported framework. Specific features in this release include:

• Parallel execution: Steps run concurrently when possible, maximizing performance
• Error handling: Distinguish between business logic failures and system errors
• State management: Built-in support for durable execution with SQLite backend

Secure Component Isolation

GenAI in the modern enterprise requires workflow isolation for security and resource control. StepFlow builds this in at the core:

• Process isolation: Each component runs in its own sandboxed process or node
• JSON-RPC protocol: Clean separation between workflow runtime and components
• Resource controls: Strict environment and resource management for each step

Open and Extensible

Workflows are defined in JSON or YAML and submitted via a wire protocol. This means that you are in complete control of all aspects of runtime execution:

• Language agnostic: Write components in Python, TypeScript, or any language
• Plugin architecture: Easy integration with existing tools and services
• MCP support: Use Model Context Protocol tools as workflow components
• Built-in components: OpenAI integration, file operations, and more out of the box
• User Defined Functions: Take control of your workflow via providing your own UDFs

Geting Started

Get the latest release from GitHub https://github.com/riptano/stepflow/releases.

Here's a simple workflow to try to get started.

# math-workflow.yaml
input_schema:
  type: object
  properties:
    numbers:
      type: array
      items:
        type: number

steps:
  - id: sum_numbers
    component: python://sum
    args:
      values: { $from: $input, path: numbers }

  - id: square_result
    component: python://square
    args:
      value: { $from: sum_numbers, path: result }

outputs:
  sum: { $from: sum_numbers, path: result }
  squared: { $from: square_result, path: result }

What's Next?

This is just the beginning! Our roadmap includes:

• Container-based components: Run components in Docker containers for ultimate isolation
• Distributed execution: Scale workflows across multiple machines
• Enhanced Python SDK: Simplified component development
• Richer component library: Pre-built components for common tasks
• Kubernetes integration: Native deployment to cloud environments

Join the Community

StepFlow is open source and we welcome contributions! Whether you want to:

• Build new components
• Improve documentation
• Report bugs or suggest features
• Share your workflows

Visit our GitHub repository to get involved.

Acknowledgments

Special thanks to all our early contributors and testers who helped shape StepFlow. Your feedback and contributions have been invaluable in getting us to this first release.

Next Steps

Ready to build your first GenAI workflow? Check out our documentation and join us in making AI workflows accessible, reliable, and secure for everyone.

Happy workflow building!

The StepFlow Team