> For the complete documentation index, see [llms.txt](https://doc.aissist.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.aissist.io/tutorial/try.md). # Try Test your Aissist workspace in Try before going live - set input context, send messages, read the reply and what shaped it, and fix common issues. *Last updated: October 25, 2026* **Try** (formerly the Simulator) is where you test your workspace before customers see it. It shows each reply together with what shaped it: the sub-agents activated, the actions called, the sources used, and any escalation or summary. ### Start a test conversation 1. Go to **Try** in the sidebar. It tests the workspace selected at the top of the console. 2. If the workflow depends on who the customer is, set the input context first (see below). 3. Click **Start**. 4. Type a message in **Type a message...** and send it. To test images or videos, use the upload button next to the message box. To start over with a fresh conversation, click the restart button next to the send button. ### Set input context When deployed, your help desk sends information about the customer with each conversation, such as their email or order number. In **Try**, you supply it yourself. Click the settings button next to the message box to open **Set Input Context**. Enter the information in **Input Context**, for example: ``` email: jane@example.com order number: E12345678 ``` Click **OK**. It is sent with every message of the test, so you can test order lookups and other customer-specific workflows as they will run live. ### Read the reply Each reply can come with: * **Reasoning** and **Sources**: why the agent answered as it did, and the passages it used. Open **Sources** to see each one. * **Sub-agents**: the sub-agents activated, with the reason. Each becomes a tag in your help desk when deployed. * **Actions**: each action called and the parameters sent. * **Search**: searches the agent ran. * **Escalation**: the handoff details, when the conversation escalated. It becomes an internal note in your help desk. * **Summary** and **Comment**: structured notes that also become internal notes in your help desk. ### Inspect the session The panel on the right shows the session as a whole: * The **Engine**, the **State**, the **Primary Language** and **Detected Language**, whether **Respond Only to Tagged?** is on, and the **Active Subagents**. * **Session**: how long the reply took, step by step, and every **Sub-agent**, **Action**, **Source**, **Escalation**, **Summary** and **Follow-up** of the conversation. * **States**: the conversation's state values. * **Contexts**: the context the agent retrieved to write the reply, such as asset passages and action results. Use **Search contexts...** to find a word, such as a price or a policy name, and see whether the agent had it. ### What to test Use real examples from your help desk: * the most common questions * edge cases and sensitive requests * messages that should activate each sub-agent, and messages that should not * messages that should call each action, with realistic input context * requests for a person, to check escalation and handover ### Fix common issues | What you see | Where to fix it | | -------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------- | | The answer is wrong or missing, and **Contexts** does not contain the fact | Add or fix the knowledge: [Add Assets](/tutorial/add-assets.md). Check retrieval in the asset debugger. | | **Contexts** has the fact, but the answer ignores it or contradicts it | Remove conflicting content, or make the instruction clearer. | | The wrong sub-agent activates, or none does | Sharpen the sub-agent's **Scenario Description**: [Add Sub Agents](/tutorial/add-sub-agents.md). | | An action does not run, or gets the wrong parameters | Fix its trigger scenario or parameter descriptions, then check it in the action debugger: [Add Actions](/tutorial/add-actions.md). | | Tone, length or format is off | Adjust **Character** or **Instruction**: [Create a Workspace](/tutorial/create-workspace.md). | | It escalates too early or too late | Adjust the **Escalation** tab and **Escalate Immediately on Request**. | After each fix, restart and send the same messages again. ### Do it with Ask Engine Click **Ask about this session** at the top of **Try** to ask Ask Engine about the conversation you just had: > Why did it say the refund takes 10 days? Where did that come from? > Why didn't the order tracking sub-agent activate on my second message? Ask Engine reads what the agent was working from at each turn - the sub-agent, its instruction, the sources and action results - and explains the reply. To teach the agent a better answer, ask Ask Engine to add a continuous learn example: it shows you the change first, and saves it only when you confirm. See [Ask Engine](/capabilities/ask.md) and [Continuous Improvement](/capabilities/continuous-improvement.md). ### Best practice * Test one workflow at a time, with the customer's real wording. * Test again after every change to instructions, assets, sub-agents or actions. * Fix the cause at its source, then rerun the same messages until the result is stable. * Keep a short list of test messages per workflow and rerun it before each release. For a repeatable test set, see [Evaluation](/capabilities/evaluation.md). ### Related * [Deploy](/tutorial/deploy.md) * [Add Assets](/tutorial/add-assets.md) * [Add Actions](/tutorial/add-actions.md)