Low-level BEAM↔Python concurrency substrate.

Slither provides three reusable primitives on top of SnakeBridge/Snakepit:

• Store — ETS-backed shared state with lock-free concurrent reads and serialized writes, exposed to Python via bidirectional tool calls.

• Dispatch — Bounded batched fan-out to Python pools with real backpressure, ordering control, and policy-driven error handling.

• Pipe — Stage composition over BEAM + Python steps with routing, telemetry, and supervised execution.

Quick Start

# 1. Define a pipe
defmodule MyPipe do
  use Slither.Pipe

  pipe :my_pipe do
    stage :transform, :beam, handler: &MyApp.transform/2
    stage :predict, :python,
      module: "model", function: "predict",
      pool: :gpu, batch_size: 32
    output :default
  end
end

# 2. Run it
{:ok, results} = Slither.Pipe.Runner.run(MyPipe, input_data)

Architecture

Slither adds exactly three things on top of SnakeBridge/Snakepit:

• Shared state on BEAM with Python-visible views
• Bounded batching + backpressure
• Stage composition + routing

Everything else — Python env management, pool lifecycle, retries, circuit breakers, streaming mechanics — is delegated to SnakeBridge/Snakepit.