Simulation
Simulation
Check how your agent answers before real users see it: write scenarios, run them on Published or Draft, read the results, and keep tools safe with Tool Mocks and Tool Policy.
Simulation lets you check how your agent answers before real users see it. You write test conversations (scenarios) with clear pass criteria, run them against the Published or Draft version of your agent, and read which answers matched. In the CMS, all of this lives under Simulation in the agent sidebar.
Suggested criteria, the agent's answers and the match outcomes are produced by AI, so they can differ from one run to the next.
The pages follow the order you work in: get started, build your tests, run and review them, then handle agents that use tools.
Get started
| Page | What you'll learn |
|---|---|
| Overview | What Simulation is for, the key terms, who can use it and where to find it. |
| Your First Simulation | Go from an agent with no tests to a read result: make a folder, write a two-step scenario, run it and read the outcome. |
Build your tests
| Page | What you'll learn |
|---|---|
| Organize Scenarios in Folders | Group scenarios into folders and subfolders, move and duplicate them, and run a whole folder at once. |
| Write Good Scenarios | Write multi-step conversations and criteria that are specific enough to tell a good answer from a weak one. |
Run and review
| Page | What you'll learn |
|---|---|
| Run Simulations | Choose between Published and Draft, start a run and follow it. |
| Read the Results | Filter the Results tab and read each step, its criteria and its tool calls. |
| Fix and Retest | Turn a failed criterion into a change on your agent, retest it on Draft and publish when it passes. |
Agents with tools
| Page | What you'll learn |
|---|---|
| Tools in Simulations | What happens when your agent calls a tool during a test, and how Qlar decides whether the call is mocked, simulated, live or blocked. |
| Tool Mocks | Script the answers your tools give, so every run gets the same predictable results. |
| Tool Policy | Decide what each tool does in simulations when no mock answers the call. |
Also in the docs
- A guided example for Atlas: Test Your Agent's Quality continues the Kirana Group story and builds a test suite as part of setting up an agent.
- Field-by-field reference: Simulation in Help explains every button, field and status on the Scenarios, Results, Tool Mocks and Tool Policy screens.
Tip: New to Simulation? Start with the Overview, then follow Your First Simulation. It takes about ten minutes.