Skip to content

The Framework for AI Systems That Learn

Go beyond static prompts. Flujo provides the structure, state, and observability to build AI agents that analyze their own performance, adapt over time, and improve with every run.

Key Features

  • Compose pipelines using a simple DSL
  • Share and validate state with Pydantic models
  • Integrate loops, branching, and human-in-the-loop patterns
  • Agent infrastructure with factory functions
  • Centralized prompt management
  • Enhanced serialization with global custom serializer registry
  • Rich internal tracing and visualization (FSD-12)
  • Integrated cost and token usage tracking

Getting Started

Project-Based Workflow

Start new workflows as self-contained projects and run everything from the project root without passing file paths.

  • Initialize a project: flujo init
  • Create a pipeline conversationally: flujo create
  • Run the pipeline (project-aware): flujo run --input "..."
  • Inspect and replay runs: flujo lens list, flujo lens trace <run_id>, flujo lens replay <run_id>

See the Quickstart and the Architect guide for details.

Advanced Features

Use the navigation on the left to explore the guides, examples, and API reference.