examples

Examples

A tutorial-ordered introduction to wfpy's dataflow programming model. Each file introduces one idea, is runnable, and is executed by tests/test_examples.py on every test run — so an example that stops working fails CI rather than rotting.

Read docs/dataflow-concepts.md alongside these; the examples use its vocabulary (actor, token, firing, guard) without re-explaining it.

Every example runs two ways:

python examples/01_simple_task.py            # runs it, prints the result
wfpy run examples/01_simple_task.py --input In=21   # same graph, via the CLI
wfpy plan examples/01_simple_task.py --format graph # the graph as JSON

Nothing here needs credentials, a network, or an optional dependency — the one agent example (09) uses the offline transport="mock".

The dataflow model

Plain @task actors, so the semantics stay in the foreground. Everything here applies unchanged to agents, because an @agent is an actor (09).

	Example	Introduces
01	01_simple_task.py	Ports, @action(consumes=, produces=), tokens, firings
02	02_pipeline.py	Chaining tasks; parameters (annotated field, no default)
03	03_streams_and_nondeterminism.py	Several actions per task; loop() streams; nondeterminism
04	04_guarded_actions.py	@guard — firing conditions on values and state
05	05_state.py	State fields (annotated field with a default)
06	06_schedules.py	class Schedule — an FSM sequencing a task's actions
07	07_priorities.py	class Priority — deterministic action ordering
08	08_networks.py	Sources, sinks, fan-out, and self-terminating feedback
09	09_agents_are_actors.py	An @agent is a task; the offline mock transport

Agent orchestration

The patterns you reach for when wiring LLMs together. Each runs offline via transport="mock" and is CI-tested; swap in a real transport unchanged.

	Example	Pattern
10	10_parallel_agents.py	Fan out to parallel specialists, reduce with a task (map-reduce)
11	11_routing.py	Dispatch each request to the right specialist (guard routing)
12	12_repair_loop.py	Generate → check → revise until good (feedback loop)

Reading order

01 → 02 establish tasks, ports, and wiring. 03 introduces multiple actions (and the nondeterminism that motivates 06 and 07). 04 adds guards, 05 adds state — the two things that make an actor more than a function. 06 (schedules) and 07 (priorities) are the two ways to make action selection deterministic. 08 is the network-level view: sources, sinks, fan-out, feedback. 09 is the payoff — an LLM agent slotted into the same model. 10–12 then apply that model to multi-agent orchestration: fan out and reduce, route, and loop.