> This page is for Omni Channel. > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.hams.ai/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.hams.ai/_mcp/server. # Playground > Test an AI agent in a chat or on a call before it answers customers. The playground is a test conversation with the AI agent that no customer sees. Use it to find the points where the AI agent does not follow your instructions. You can test the chat settings with a test chat. You can test the voice settings with a test call from your browser. > **Note** > > The AI agent is off by default. Hams support turns on the **Workflows + Agent** feature for your account. Only administrators can see **AI** in the sidebar and use the playground. Messages in the test chat do not use AI credits, and the AI agent report does not count them. ![The Test chat panel: your messages go to the AI agent, but not to a customer.](/_fern-img/34978fedcae406f3e85d40a4dde018a277e5e9522e637273a36dbbd1764322a1.webp) ## Open the playground Open a saved AI agent. In the header of the editor, click **Test**. The panel opens on the right side of the editor. The panel shows only on a wide screen. The switch at the top of the editor selects the panel: * With **Chat** selected, the panel is **Test chat**. See [Send a test message](#send-a-test-message). * With **Voice** selected, the panel is **Test call**. See [Test a voice agent](#test-a-voice-agent). The panel stays open while you move between the tabs of the editor. To close it, click the close icon, or click **Test** again. A new AI agent does not show **Test** until you save it. ## Send a test message 1. Type a message in **Type your message...**. 2. Press Enter, or click the send button. Each reply shows the name and the avatar of the AI agent. The AI agent replies with its last saved instructions, knowledge, tools, and scenarios. If the editor has unsaved changes, the panel shows **Uses the last saved version**. Save the AI agent after a change. Then send a new message to test the change. If no reply comes in 2 minutes, the panel shows **The agent could not reply. Try again.** ## What each reply shows Under a reply, the playground can show 2 lists: * **Sources**: the documents that gave the facts for the reply. * **Actions**: the tools that the AI agent used for the reply. A transfer to a scenario also shows here. Click a tool in **Actions** to see its details under **Tool info**: | Item | What it shows | | ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Input** | The values that the AI agent gave to the tool. | | **Request sent** | For a custom tool: the method, the URL, the headers, and the body of the request. The system hides the credentials and the values of the static headers. | | **Response** | The response that the tool got, with the HTTP status code if there is one. **Truncated** shows if the response is too long to show in full. | ## How the playground is different from a conversation A playground message does not belong to a real conversation or contact. Because of this: * The tools that change a conversation or a contact do nothing. These are **Add Private Note**, **Add Contact Note**, **Update Priority**, **Add Label to Conversation**, **Update Contact Field**, and **Collect CSAT**. The tool shows in **Actions**, but no record changes. * **Handoff to Human** and **Resolve Conversation** show in **Actions** only. No conversation moves to a human agent, and no conversation closes. * The variables in the instructions, for example `{{ contact.name }}`, are empty. * A custom tool with the type **Send files** sends no files. The playground shows a preview of each file under the reply. Images, audio, and PDF files show in the playground. Each file has a link. > **Warning** > > Do not test a custom tool that changes data on a live system unless you accept the change. A custom tool with the type **HTTP request** sends a real request to your endpoint from the playground. ## Keep or clear the conversation The playground keeps the conversation of each AI agent while you move to other tabs. Then you can change the instructions and come back to the same conversation. When you reload the page, the conversation starts again. To clear the conversation, click the reset icon in the header of the **Test chat** panel. ## Test a voice agent The **Test call** panel connects your browser to the voice agent. You talk to the AI agent through your microphone and hear its voice. The call uses the last saved voice settings. It does not use a phone line. Before you start, make sure of these conditions: * **Answer calls with this agent** is on, and you saved the AI agent. See [Voice agents](/cloud-call-center/voice-agents). * The browser has permission to use your microphone. **Start call** is not available when the Hams installation has no setup for test calls. In that case, contact Hams support. #### Open the test call At the top of the editor, click **Voice**. Then click **Test**. The **Test call** panel opens with **Call your agent**. #### Start the call Click **Start call**. If the browser asks for the microphone, give permission. The panel shows **Connecting…**. #### Talk to the AI agent When the call connects, the AI agent opens the call as set in **Who speaks first**. Speak to the AI agent. #### End the call Click **Hang up**. During the call, the panel shows a transcript of what you say and what the AI agent says. A line under the transcript shows the state of the AI agent: **Listening**, **Thinking**, or **Speaking**. Use these controls during the call: | Control | What it does | | -------------------------------- | ----------------------------------------------------------------------------- | | **Mute** / **Unmute** | Turns your microphone off and on. | | **Type a message to the agent…** | Sends text to the AI agent instead of speech. Press Enter, or click **Send**. | | **Hang up** | Ends the call. | The call also ends when the AI agent hangs up, or when you close the panel. It ends when you click **Chat** at the top of the editor too. After a call, click **Call again** to start a new call. A new call clears the transcript. If the call cannot start, the panel shows a message: | Message | What to do | | ------------------------------------------------------------- | ----------------------------------------------------------------------------------------------- | | **Turn voice on and save before calling the agent.** | Turn on **Answer calls with this agent**. Then click **Save**. | | **Allow microphone access to call the agent.** | Give the site permission to use the microphone in the browser. Then click **Start call** again. | | **No voice agent is free right now. Try again in a moment.** | Wait, and then start the call again. | | **Too many test calls started. Wait a minute and try again.** | Wait 1 minute, and then start the call again. | | **The call couldn't be connected. Try again.** | Start the call again. If the call does not connect in 30 seconds, this message also shows. | ## What to test Test these cases before you assign the AI agent to customers: * The most common question of your customers. Examine the answer and the **Sources**. * A question that the AI agent must not answer. Make sure that it follows the **Guardrails**. * A request that needs a human agent. Make sure that the AI agent uses **Handoff to Human**. * A request that needs a tool. Make sure that the AI agent uses the correct tool with the correct **Input**. * A request for each scenario. Make sure that the correct scenario shows in **Actions**. * A question that the knowledge does not answer. Make sure that the AI agent does not invent an answer. * For a voice agent, a test call. Listen to the greeting, the voice, and the pauses between turns. Interrupt the AI agent to test **Turn taking**. After the AI agent answers customers, administrators can see how the AI agent wrote each real reply. See [Assistants](/omni-channel/ai-agent/assistants#see-how-a-reply-was-generated). ## Related * [Assistants](/omni-channel/ai-agent/assistants) * [Tools](/omni-channel/ai-agent/tools) * [Scenarios](/omni-channel/ai-agent/scenarios) * [Knowledge and documents](/omni-channel/ai-agent/knowledge)