First workflow in five minutes

Before you start

Complete the Python install section (maturin develop in sdk-python/). The example below is Python. The TypeScript equivalent lives in TypeScript quickstart.

You do not need an LLM API key, Docker, or a running server for this path.

Concept

This is the shortest path to a working ArcFlow run. You define two agents, chain them in order, call run(), and read the final output. With no provider= argument, the runtime uses its built-in default agent backend, which returns deterministic placeholder text so you can validate wiring before connecting OpenAI or another model.

An Agent is a named role with instructions. A Workflow is an ordered list of agents executed by the Rust runtime. Python declares structure; it does not execute LLM calls itself.

Example

Save as first_workflow.py anywhere on your machine (with the arcflow package importable from your virtual environment):

python

from arcflow import Agent, Workflow

researcher = Agent(
 name="researcher",
 role="research",
 instructions="Research the given topic thoroughly.",
)
writer = Agent(
 name="writer",
 role="write",
 instructions="Write a clear summary of the research.",
)

workflow = Workflow("research_pipeline")
workflow.step(researcher)
workflow.step(writer)

result = workflow.run("Analyze renewable energy trends")
print(result.output)
print(f"run_id={result.run_id} steps={result.step_count} status={result.status}")

Run it:

bash

python first_workflow.py

Verify

Check	Expected
Script exits without exception	Yes
First `print` line	Non-empty text
`result.step_count`	`2`
`result.status`	`completed`
`result.run_id`	UUID string

Exact output text may vary by runtime version. The integration test asserts len(result.output) > 0 and result.step_count == 2.

This matches the canonical example in sdk-python/tests/integration/test_first_five_minutes.py.

What just happened

You declared two agents and registered them as ordered steps on a Workflow. Calling run() serialized the definition and input to the ArcFlow Rust runtime, which executed step one then step two. The last step's output surface is result.output. Trace metadata is available on result.trace_events and via workflow.trace() after the run.