Choose an executor

Avalanche schedules a workflow topologically. The executor decides where node functions run; it does not change the workflow’s dependency graph or provider binding semantics.

​

Local execution

Use ava.LocalExecutor for direct local execution:

run = my_workflow().run(executor=ava.LocalExecutor())
result = run.result()

The local executor invokes node functions in the local coordinator process.

​

Ray execution

Install the optional Ray extra, then select ava.RayExecutor in embedded mode or pass --ray to the local operator:

uv add "avalanche-ai[ray]"
uv run ava dev --flows ./flows --ray

run = my_workflow().run(executor=ava.RayExecutor())
result = run.result()

With Ray, object references remain distributed until a consumer needs materialization. The DAG runtime controls dependency order; the local operator and UI remain control-plane clients rather than the execution engine.

​

Select deliberately

• Choose LocalExecutor for the normal local development path and simple embedded workflows.
• Choose RayExecutor only when a local Ray runtime is installed and the workflow should execute through Ray.

Ray is optional and participates only when RayExecutor is selected.