A durable workflow and AI agent orchestration engine that persists every step to survive crashes and failures. It separates orchestration from business logic so the workflow graph remains deterministic by construction while workers run as plain code.
Project overview
The engine separates orchestration from business logic to achieve architectural determinism and is documented as scaling horizontally to billions of workflow executions.
Project type
AI Agent · Workflow
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Developers and enterprise teams building distributed applications or AI agent pipelines who need infrastructure that automatically handles failures, retries, and state recovery.
Key capabilities
Every step is persisted, surviving crashes, restarts, and network failures with configurable retries and timeouts.
Orchestration is separated from business logic so determinism is architectural; workers run any code and the workflow graph stays deterministic by construction.
Orchestrates AI agents with native integrations for 14+ LLM providers, function calling, human-in-the-loop approval, and vector databases for RAG.
Supports workers written in Java, Python, Go, JavaScript, C#, Ruby, or Rust without SDK lock-in or determinism constraints.
Supports dynamic forks, tasks, and sub-workflows resolved at runtime, enabling LLMs to generate and execute JSON workflow definitions immediately.
Restart workflows from the beginning, rerun from any task, or retry just the failed step, preserving full execution history indefinitely.
Scales horizontally to billions of workflow executions, battle-tested at Netflix, Tesla, LinkedIn, and J.P. Morgan.
Provides a visual UI for designing, running, and debugging workflows with full observability of inputs, outputs, timing, and retry history.
Limitations and risks
Running the CLI quick start path requires Node.js v16+ and Java 21+ to be installed on the system.
Building the framework from source requires Docker Desktop, Java (JDK) 21+, Node.js 18+, and pnpm.
Getting started
The setup provides a 60-second start guide using the CLI or Docker. For the CLI path, install the package globally via npm and then access the visual UI at localhost:8080.
Alternatives and comparisons
A source-available platform to build and deploy AI agents and workflows using a visual canvas, custom code, and a large library of integrations.
A visual authoring tool for AI agents and workflows that includes an interactive playground and built-in API capabilities.
A platform featuring a low-code interface to build, deploy, and manage continuous AI agents via a server component.