F#AI workflowsPipeline engineMITUpdated July 15, 2026

View the repository

F#kYeah implements the StrongDM Attractor specifications as a DOT-driven pipeline runner. A graph describes stages, transitions, conditions, retries, human gates, and tool execution; the engine runs that graph against multiple LLM providers and checkpoints progress so interrupted workflows can resume.

DOT is a strong fit because it keeps control flow declarative and inspectable. A workflow can be rendered as a diagram, validated before execution, and versioned as text. F# adds a second layer of clarity: discriminated unions and pattern matching are natural tools for representing stage types and execution outcomes that should be handled exhaustively.

Why it is worth exploring

  • It models agent workflows as inspectable programs with explicit control flow.
  • Simulation, validation, checkpoints, and conformance tests make workflow behavior testable without spending model tokens on every iteration.
  • The unified model client separates pipeline semantics from any single provider.

Start with a small graph under examples, use the schema command to understand its attributes, and then follow graph parsing into the execution engine.

Updated: