Slither.Pipe (Slither v0.1.0)

Compose BEAM and Python stages into a supervised pipeline.

A Pipe is a sequence of stages with routing, error handling, and session-scoped tool registration. Define pipes using the DSL and execute them with Slither.Pipe.Runner.

Example

defmodule MyPipe do
  use Slither.Pipe

  pipe :my_pipe do
    store :features, MyApp.FeatureStore

    stage :enrich, :beam,
      handler: &MyApp.enrich/2

    stage :score, :python,
      executor: Slither.Dispatch.Executors.SnakeBridge,
      module: "my_model",
      function: "predict_batch",
      pool: :gpu_pool,
      batch_size: 32

    stage :route, :router,
      routes: [
        {&MyApp.high_confidence?/1, :accept},
        {&MyApp.low_confidence?/1, :review},
        {fn _ -> true end, :reject}
      ]

    output :accept
    output :review
    output :reject

    on_error :score, :skip
    on_error :*, :halt
  end
end

# Run the pipe
{:ok, results} = Slither.Pipe.Runner.run(MyPipe, input_data)
# => %{accept: [...], review: [...], reject: [...], default: [...]}