Skip to main content

​
Why SuperDialog flows are easy to test

SuperDialog is text in, text out. There is no audio to mock, no WebRTC room to spin up, no telephony to stub. Every dialog is a Python function that takes a string and returns a string. This is the killer feature vs. voice-coupled frameworks where tests need audio fixtures.

​
Setup

pip install pytest pytest-asyncio
Configure pytest-asyncio in pyproject.toml: 
[tool.pytest.ini_options]
asyncio_mode = "auto"
Or use anyio as specified in the project guidelines.

​
Basic test

import pytest
from superdialog import DialogMachine, Flow

@pytest.mark.asyncio
async def test_greets_customer():
    dm = DialogMachine(
        flow=Flow.load("kyc.json"),
        llm="anthropic/claude-haiku-4-5",
    )
    reply = await dm.turn("Hello")
    assert reply.text  # non-empty response

​
Testing slot collection

Use machine.state["slots"] to assert that the flow collected the right data: 
@pytest.mark.asyncio
async def test_kyc_collects_aadhaar():
    dm = DialogMachine(
        flow=Flow.load("kyc.json"),
        llm="anthropic/claude-haiku-4-5",
    )

    await dm.turn("I need to verify my KYC.")
    reply = await dm.turn("My Aadhaar ends in 1234.")

    assert "1234" in reply.text or dm.state["slots"].get("aadhaar_last_4") == "1234"

​
Testing multi-turn flows

@pytest.mark.asyncio
async def test_appointment_confirmation_flow():
    dm = DialogMachine(
        flow=Flow.load("appointment.json"),
        llm="anthropic/claude-haiku-4-5",
    )

    r1 = await dm.turn("I'm calling about my appointment.")
    assert any(word in r1.text.lower() for word in ["friday", "time", "appointment"])

    r2 = await dm.turn("Friday 4pm works for me.")
    assert any(word in r2.text.lower() for word in ["confirmed", "great", "see you"])

​
Testing with mock tools

from superdialog import PythonTool

@pytest.mark.asyncio
async def test_tool_is_called():
    calls = []

    def lookup_customer(customer_id: str) -> dict:
        """Look up customer by ID."""
        calls.append(customer_id)
        return {"name": "Ravi Kumar", "verified": True}

    dm = DialogMachine(
        flow=Flow.load("kyc.json"),
        llm="anthropic/claude-haiku-4-5",
        tools=[PythonTool.of(lookup_customer)],
    )

    await dm.turn("My customer ID is CUST-999.")
    assert "CUST-999" in calls

​
Testing flow reset

@pytest.mark.asyncio
async def test_reset_clears_state():
    dm = DialogMachine(flow=Flow.load("kyc.json"), llm="anthropic/claude-haiku-4-5")

    await dm.turn("My Aadhaar ends in 1234.")
    assert dm.state["slots"].get("aadhaar_last_4") == "1234"

    dm.reset()
    assert not dm.state["slots"].get("aadhaar_last_4")

​
Testing flow switching

from superdialog import FlowSet

@pytest.mark.asyncio
async def test_switch_to_escalation():
    flowset = FlowSet({"main": main_flow, "escalation": escalation_flow})
    dm = DialogMachine(flow=flowset, llm="anthropic/claude-haiku-4-5")

    dm.switch_flow("escalation")
    reply = await dm.turn("I want to speak to a manager.")
    assert "escalat" in reply.text.lower() or "manager" in reply.text.lower()

​
Traversal snapshots for eval

Set traversal_dir on your test machine to capture every conversation as a JSON file. Use these files as your eval corpus: 
@pytest.mark.asyncio
async def test_and_capture_traversal(tmp_path):
    dm = DialogMachine(
        flow=Flow.load("kyc.json"),
        llm="anthropic/claude-haiku-4-5",
        traversal_dir=str(tmp_path),
    )

    await dm.turn("Start KYC")
    await dm.turn("1234")
    await dm.turn("1990-05-15")

    # When dm reaches terminal node, a JSON file is written to tmp_path
    snapshots = list(tmp_path.glob("*.json"))
    assert len(snapshots) == 1

​
Tips

  • Use a cheap model (claude-haiku-4-5) in tests to keep costs low and latency fast
  • Use dm.reset() between test cases if reusing a machine instance
  • Keep flow JSON files in version control so tests are deterministic
  • Build a corpus from traversal snapshots, then use superdialog eval (v0.3) to run regression tests