Introduction to ToolSimulator
ToolSimulator is a tool simulation framework that enables efficient and scalable training of tool-augmented LLM. It is designed to test AI agents that rely on external tools, at scale. ToolSimulator uses the tool schema, the agent’s input, and the current simulation state to produce a realistic, context-appropriate response.
Setting up ToolSimulator
To set up ToolSimulator, you need to decorate and register your tools. This involves defining the tool schema and the agent’s input. ToolSimulator then uses this information to produce a realistic response. You can also steer the simulation with context, allowing for more complex and realistic testing scenarios.
Configuring Stateful Tool Simulations
ToolSimulator allows for stateful tool simulations, which enables testing of multi-turn agent workflows. This involves maintaining consistent shared state across tool calls, making it safe to test database interactions and multi-step processes without touching production systems. You can also enforce response schemas with Pydantic models, ensuring that the responses from the tool simulations are consistent and valid.

Integrating ToolSimulator into a Complete Evaluation Pipeline
ToolSimulator can be integrated into a complete Strands Evals evaluation pipeline, allowing for thorough and efficient testing of AI agents. This involves setting up the tool simulations, configuring the stateful tool simulations, and enforcing response schemas. You can also apply best practices for simulation-based agent evaluation, ensuring that the testing is thorough and reliable.
How this compares
How this compares
| Component | Open / This Approach | Proprietary Alternative |
|---|---|---|
| Model provider | Any — OpenAI, Anthropic, Ollama | Single vendor lock-in |
🔑 Key Takeaway
The use of ToolSimulator enables efficient and scalable testing of AI agents, enhancing their reliability and performance. By using ToolSimulator, you can thoroughly and safely test AI agents that rely on external tools, at scale.
Key Links