Examples

The repository’s standalone examples are smoke-tested. Run them from the examples/ directory with the local operator and browser UI:

cd examples
uv run ava dev

​

Start here

​

complex_dag_pattern.py

The smallest local execution path. It demonstrates the Python DAG API, explicit data passing, fan-out, fan-in, and ava.LocalExecutor without requiring Iceberg, Ray, the operator, or the TUI.

​

operator_workflow.py

A flow module for the local operator and connected UI path.

​

Data workflow patterns

​

stream_pattern.py

Incremental processing with local Iceberg tables and ava.Stream in append_scan mode. Each consumer edge receives a unique stream key.

​

cursor_pattern.py

Manual checkpoint control with ava.Cursor, including a source-table snapshot, model-specific cursors, and a multi-table synchronization checkpoint.

​

document_file_workflow.py

Typed ava.File inputs and outputs through a BaseInput workflow.

​

Flagship walkthrough: customer feedback review

customer_feedback_review/ is a production-shaped example of an Avalanche workflow. It:

1. loads a typed customer-feedback workbook;
2. runs theme and risk analysis agents in parallel;
3. reconciles the reports deterministically and rejects inconsistencies;
4. renders an Excel review workbook and Word executive brief in parallel;
5. publishes the validated artifacts to an output directory.

Start it against its dedicated flow module:

uv run ava dev --flows customer_feedback_review/flow.py

Generated example artifacts land beneath .avalanche/outputs/feedback_review/ by default. Set AVALANCHE_EXAMPLE_ROOT to isolate example artifacts.

The example is deliberately richer than the starter workflow. Use it after you understand the source → agent step → deterministic action shape from Option A — Starter project (recommended).