> ## Documentation Index > Fetch the complete documentation index at: https://docs.get3rd.com/llms.txt > Use this file to discover all available pages before exploring further. # Simulator > Watch an AI agent try to complete a real task on your website, step by step — and see what stopped or slowed it. The **Simulator** gives an AI agent a task on your own website, such as finding a price or signing up, and lets you watch it try. You see every step it takes, whether it got there, and the problems it ran into along the way. It lives under **Agent readiness**, in the **Simulator** tab. Where the [Agent readiness](https://docs.get3rd.com/drivers/agent-readiness.md) tests check your site against a list of rules, a simulation shows you what an agent actually experiences. ## Run a simulation 1. Go to **Agent readiness** and open the **Simulator** tab. 2. Click **Add simulation**. 3. Write the **Instructions** — what the agent should do, as you'd ask a person. For example: *Find and add the Adidas Samba OG in size 42 to the basket.* Or pick one of the suggestions below the box. 4. Check the **Website**. It starts on your brand's site. You can type another site your brand tracks, such as a shop or app subdomain. 5. Check the **Type of site** and the **Market**. The type decides which suggestions you see and how the run is judged. The market decides which country the agent browses from. 6. Click **Run**. Every simulation runs as **ChatGPT**, in a real browser. > [!NOTE] > > Simulations run only on the sites your brand tracks. To simulate on another site of yours, add it under [Monitoring settings](https://docs.get3rd.com/workspace/monitoring-settings.md) first. ## A run keeps going Starting a simulation files it and returns you to the run. Closing the tab, or leaving the page, does not stop it. Open the same simulation again and the run picks up where it is, including the steps already taken. **Stop** ends the run. It works from any tab that has the simulation open. ## Watch it live While the simulation runs, you see the agent's browser live, next to the list of steps it has taken. Click a step to see the page as it was at that moment. Sometimes the agent needs something from you: * **Your details** — a sign-up form may ask for a name, email or password. Your answer is encrypted in your browser and typed straight into the site. The agent never sees it, and nothing is saved. You can **Skip** instead. * **Your approval** — before a step that commits to something, such as paying, checking out, creating an account, sending a message or sharing personal data, the agent stops and asks. Only the person who started the simulation is asked. If your site offers a shopping route or an MCP server for agents, the agent uses it instead of browsing, and the run page says so. ## Read the results When the simulation finishes, the run page shows: * **Results** — how the run ended, what happened in one line, how many **issues to fix** it found and how many checks are **working**. * **Insights** — one card per finding, worst first. A card shows the stage it belongs to, its impact, what the agent ran into and on which step. * **Simulation** — every action the agent took, numbered, and the page it landed on. * **Replay** — a recording of the whole run. Each card that has something to fix has an **Add task** button. If the finding rests on an Agent readiness test, **Open test** takes you to it. ## The simulations list The **Simulator** tab lists every simulation for your brand, newest first, with its website, market, result, number of issues and duration. Filter by market, website, result or date, open a simulation to see its run page, or delete one. A result reads **Queued**, **Running**, **Success**, **Not completed**, **Stopped**, or **Failed**. **Queued** means the run is waiting for a free slot. **Failed** means the run crashed or timed out. **Stopped** means someone pressed **Stop**. ## Simulations on the Issues tab The serious problems from each site's latest simulation also appear on the Agent readiness **Issues** tab, marked **Found by a simulation**. Each one opens the simulation it came from and can be added as a task. A few problems stay off the Issues tab: * Problems an Agent readiness test already measures, such as bot protection. The test's own issue covers them. * Minor problems, just as minor test findings stay on the **Tests** tab. A newer simulation on the same site replaces the older one's problems. ## Tasks from simulations A problem becomes one task per site, whether you add it from the run page or the Issues tab. If a later simulation runs into the same problem, it points to the same task rather than creating a new one. Simulation tasks appear in [Task management](https://docs.get3rd.com/actions/task-management.md). They aren't added to [playbooks](https://docs.get3rd.com/actions/playbooks.md) automatically, because one run is a single observation rather than a measurement of your whole site. ## Limits * At most one simulation can be pending or running per brand. A second start shows **A simulation is already running for this brand. Wait for it to finish before starting another.** * Up to 20 starts per brand in any 24 hours. * Your organisation runs two simulations at a time. Any further run waits in the queue. ## Related * [Agent readiness](https://docs.get3rd.com/drivers/agent-readiness.md) — the tests a simulation sits beside * [Task management](https://docs.get3rd.com/actions/task-management.md) — where simulation findings become work This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.