Simulators

Overview

Simulators evaluate your agent across a full interaction instead of a single input/output pair. They drive multi-turn conversations and generate tool responses at runtime, so you can test how the agent handles a dialogue and works toward a goal. Strands Evals ships two: ActorSimulator plays a simulated user (or any other conversational participant), and ToolSimulator stands in for the tools the agent calls.

Why simulators?

Traditional evaluation approaches have limitations when assessing conversational agents:

Static Evaluators:

• Evaluate single input/output pairs
• Cannot test multi-turn conversation flow
• Miss context-dependent behaviors
• Don’t capture goal-oriented interactions

Simulators:

• Generate dynamic, multi-turn conversations
• Adapt responses based on agent behavior
• Test goal completion in realistic scenarios
• Evaluate conversation flow and context maintenance
• Enable testing without predefined scripts
• Simulate tool behavior without live infrastructure

When to use simulators

Use simulators when you need to:

• Evaluate Multi-turn Conversations: Test agents across multiple conversation turns
• Assess Goal Completion: Verify agents can achieve user objectives through dialogue
• Test Conversation Flow: Evaluate how agents handle context and follow-up questions
• Generate Diverse Interactions: Create varied conversation patterns automatically
• Evaluate Without Scripts: Test agents without predefined conversation paths
• Simulate Real Users: Generate realistic user behavior patterns
• Test Tool Usage Without Infrastructure: Evaluate agent tool-use behavior without live APIs, databases, or services

ActorSimulator

The ActorSimulator is the core simulator class in Strands Evals. An “actor” is any conversational participant: a user, a customer service representative, a domain expert, an adversarial tester, or anything else that engages in dialogue. The simulator holds an actor profile, generates responses from the conversation history, and tracks goal completion. Change the profile and system prompt and you change who the simulator plays.

User simulation

The most common use of ActorSimulator is user simulation: playing a realistic end-user interacting with your agent during evaluation. This is the primary use case the documentation covers.

See the User Simulation Guide for the full walkthrough.

Other actor types

To simulate an actor other than a user, supply a custom profile. The same class covers:

• Customer Support Representatives: Test agent-to-agent interactions
• Domain Experts: Simulate specialized-knowledge conversations
• Adversarial Actors: Test how the agent handles hostile or malformed input and edge cases
• Internal Staff: Evaluate internal tooling workflows

ToolSimulator

The ToolSimulator stands in for the tools your agent calls. Instead of executing the real function, it asks an LLM to generate a schema-valid response and keeps state across calls so related tools stay consistent.

Reach for it when the real tools require live infrastructure, when you need controllable behavior for evaluation, or when the tools are still under development.

from typing import Any
from pydantic import BaseModel, Field
from strands import Agent
from strands_evals.simulation.tool_simulator import ToolSimulator

tool_simulator = ToolSimulator()

class WeatherResponse(BaseModel):
    temperature: float = Field(..., description="Temperature in Fahrenheit")
    conditions: str = Field(..., description="Weather conditions")

@tool_simulator.tool(output_schema=WeatherResponse)
def get_weather(city: str) -> dict[str, Any]:
    """Get current weather for a city."""
    pass

weather_tool = tool_simulator.get_tool("get_weather")
agent = Agent(tools=[weather_tool], callback_handler=None)
response = agent("What's the weather in Seattle?")

Key capabilities:

• Decorator-based registration with automatic metadata extraction from function signatures
• Schema-validated responses via Pydantic output models
• Shared state across related tools via share_state_id (e.g., sensor + controller operating on the same environment)
• Stateful context with initial state descriptions and bounded call history cache

See the Tool Simulation Guide for the full walkthrough.

Extensibility

The simulator framework is designed to be extensible. ActorSimulator and ToolSimulator provide general-purpose foundations, and additional specialized simulators can be built for specific evaluation patterns as needs emerge.

Simulators vs evaluators

Understanding when to use simulators versus evaluators:

Aspect	Evaluators	ActorSimulator	ToolSimulator
Role	Passive assessment	Active conversation participant	Simulated tool execution
Turns	Single turn	Multi-turn	Per tool call
Adaptation	Static criteria	Dynamic responses	Stateful responses
Use Case	Output quality	Conversation flow	Tool-use behavior
Goal	Score responses	Drive interactions	Replace infrastructure

Use Together: Simulators and evaluators complement each other. Use simulators to generate multi-turn conversations, then use evaluators to assess the quality of those interactions.

Integration with evaluators

Simulators work with trace-based evaluators:

import asyncio

from strands import Agent
from strands_evals import Case, Experiment, ActorSimulator
from strands_evals.evaluators import HelpfulnessEvaluator, GoalSuccessRateEvaluator
from strands_evals.mappers import StrandsInMemorySessionMapper
from strands_evals.telemetry import StrandsEvalsTelemetry

# Setup telemetry
telemetry = StrandsEvalsTelemetry().setup_in_memory_exporter()
memory_exporter = telemetry.in_memory_exporter

def task_function(case: Case) -> dict:
    # Create simulator to drive conversation
    simulator = ActorSimulator.from_case_for_user_simulator(
        case=case,
        max_turns=10
    )

    # Create agent to evaluate
    agent = Agent(
        trace_attributes={
            "gen_ai.conversation.id": case.session_id,
            "session.id": case.session_id
        },
        callback_handler=None
    )

    # Run multi-turn conversation
    user_message = case.input

    while simulator.has_next():
        agent_response = agent(user_message)
        turn_spans = list(memory_exporter.get_finished_spans())

        user_result = simulator.act(str(agent_response))
        user_message = str(user_result.structured_output.message)

    all_spans = memory_exporter.get_finished_spans()
    # Map to session for evaluation
    mapper = StrandsInMemorySessionMapper()
    session = mapper.map_to_session(all_spans, session_id=case.session_id)

    return {"output": str(agent_response), "trajectory": session}

# Use evaluators to assess simulated conversations
evaluators = [
    HelpfulnessEvaluator(),
    GoalSuccessRateEvaluator()
]

# Setup test cases
test_cases = [
    Case(
        input="I need to book a flight to Paris",
        metadata={"task_description": "Flight booking confirmed"}
    ),
    Case(
        input="Help me write a Python function to sort a list",
        metadata={"task_description": "Programming assistance"}
    )
]

experiment = Experiment(cases=test_cases, evaluators=evaluators)

async def main():
    report = await experiment.run_evaluations_async(task_function)

asyncio.run(main())

Best practices

1. Define clear goals

Simulators work best with well-defined objectives:

case = Case(
    input="I need to book a flight",
    metadata={
        "task_description": "Flight booked with confirmation number and email sent"
    }
)

2. Set appropriate turn limits

Balance thoroughness with efficiency:

# Simple tasks: 3-5 turns
simulator = ActorSimulator.from_case_for_user_simulator(case=case, max_turns=5)

# Complex tasks: 8-15 turns
simulator = ActorSimulator.from_case_for_user_simulator(case=case, max_turns=12)

3. Combine with multiple evaluators

Assess different aspects of simulated conversations:

evaluators = [
    HelpfulnessEvaluator(),      # User experience
    GoalSuccessRateEvaluator(),  # Task completion
    FaithfulnessEvaluator()      # Response accuracy
]

4. Log conversations for analysis

Capture conversation details for debugging:

conversation_log = []
while simulator.has_next():
    # ... conversation logic ...
    conversation_log.append({
        "turn": turn_number,
        "agent": agent_message,
        "simulator": simulator_message,
        "reasoning": simulator_reasoning
    })

Common patterns

Pattern 1: Goal completion testing

The simulator sets result.structured_output.stop = True (and its own simulator.stop flag) when the actor signals it has completed the goal. Inspect that flag rather than scanning the message text:

def test_goal_completion(case: Case) -> bool:
    simulator = ActorSimulator.from_case_for_user_simulator(case=case)
    agent = Agent(system_prompt="Your prompt")

    user_message = case.input
    while simulator.has_next():
        agent_response = agent(user_message)
        user_result = simulator.act(str(agent_response))
        user_message = str(user_result.structured_output.message)

    # has_next() is False either because the simulator stopped (goal reached)
    # or because max_turns was hit. Distinguish via stop_reason if needed.
    return user_result.structured_output.stop and (
        getattr(user_result.structured_output, "stop_reason", "") == "goal_completed"
    )

Pattern 2: Conversation flow analysis

def analyze_conversation_flow(case: Case) -> dict:
    simulator = ActorSimulator.from_case_for_user_simulator(case=case)
    agent = Agent(system_prompt="Your prompt")

    metrics = {
        "turns": 0,
        "agent_questions": 0,
        "user_clarifications": 0
    }

    user_message = case.input
    while simulator.has_next():
        agent_response = agent(user_message)
        if "?" in str(agent_response):
            metrics["agent_questions"] += 1

        user_result = simulator.act(str(agent_response))
        user_message = str(user_result.structured_output.message)
        metrics["turns"] += 1

    return metrics

Pattern 3: Comparative evaluation

def compare_agent_configurations(case: Case, configs: list) -> dict:
    results = {}

    for config in configs:
        simulator = ActorSimulator.from_case_for_user_simulator(case=case)
        agent = Agent(**config)

        # Run conversation and collect metrics
        # ... evaluation logic ...

        results[config["name"]] = metrics

    return results

Next steps

• User Simulation Guide: Simulate multi-turn user conversations
• Customizing User Simulation: Custom profiles, prompts, tools, and models
• Tool Simulation Guide: Simulate tool behavior with LLM-powered responses
• Evaluators: Combine with evaluators

Related documentation

• Quickstart Guide: Get started with Strands Evals
• Evaluators Overview: Learn about evaluators
• Experiment Generator: Generate test cases automatically

Tags

simulation9