# Test a Workflow (/docs/workflows/test-a-workflow-with-a-browser-call) > For the complete documentation index, see [llms.txt](https://docs.tandemcx.ai/llms.txt). Any page on this site can be fetched as Markdown by appending `.md` to its URL. **Test** runs the workflow you have open, as it was last saved, without routing a phone number to it or publishing it first. You can place a voice call from your browser, start a run with variables you choose, or send the workflow a first SMS or chat message. While the test runs, the canvas shows the path it takes and a run panel shows what each node did. ## Start a test Select **Automations → Workflows** in the left navigation and open a workflow. **Test** sits in the top right of the editor, next to **Publish**. On a narrow window it shrinks to an icon; hover over it to see the name. The test runs the version open in the editor, as it was last saved, whether that version is a draft or published. Unsaved changes are handled before the test starts: - **A draft with unsaved changes** is saved in place when you click **Test**. - **A published version with unsaved changes** can't be saved in place. The editor says so and opens **Save as New Draft**. Save the draft, then click **Test** again to test it. The **Test version** dialog then asks which kind of test to run: | Kind | What it takes | What it does | | ---------- | ---------------------------------------------------- | ----------------------------------------------------- | | Voice Call | Nothing | Places a call from your browser into the workflow | | Run | Starting variables, optional | Starts the workflow with those variables | | SMS | A **First Message**, and optional starting variables | Starts the workflow as if that text had arrived | | Chat | A **First Message**, and optional starting variables | Starts the workflow as if a visitor sent that message | The **Voice Call** kind runs on the same telephony path as a phone call, so Play Message, Capture DTMF, speech capture, and the AI Voice Agent node behave as they do for a caller. For the other kinds, **Add Variable** adds a **Name** and **Value** pair under **Starting Variables**. Values are passed as text, a row with no name is ignored, and two rows can't share a name (`Each variable needs a unique name.`). Choose **Start Call** or **Start Run**, and grant microphone permission if your browser asks. If a run can't start, the dialog shows **Couldn't start the test run** with the reason, and the canvas stays editable. The dialog keeps the kind, variables, and message you started with, so clicking **Test** again repeats the same test. **Cancel** clears them. ## Watch the path While a test runs, the canvas is read-only in your browser. You can pan, zoom, and click nodes, but not move, add, connect, or delete them, and undo, redo, and save are off. The editor header shows **Test run in progress** and the version under test. As the run moves: - Once the first node runs, the nodes the run hasn't reached are dimmed. - Each edge the run took is drawn thicker and animated. The other edges fade. - The node running or waiting right now pulses. - Each visited node carries a badge with its step number. A node visited more than three times shows its first three steps and a `+n` count for the rest. **Exit Test** in the header leaves the test and unlocks the canvas. Exiting during a voice test hangs up the call. A Run, SMS, or Chat test carries on to its end on the server after you exit; the editor stops showing it. ## The run panel The **Test Run** panel opens on the right when the test starts. The **Run Panel** switch in the header hides and shows it. The top of the panel says where the run is: | Status | Meaning | | ------------------------------------------- | --------------------------------------------------------------- | | Starting… | No node has run yet | | Running | The workflow is running | | Waiting for the next message at _node name_ | An SMS or chat test reached a node that waits for a reply | | Ended | The workflow finished | | Ended with an error | The workflow stopped on an error | | Caller disconnected | The workflow ended with a disconnect | | Call ended | A voice test call ended before the workflow recorded its finish | Below that, **Nodes** lists every step in order with the node's name and its state: **Running**, **Waiting**, **Done**, or **Error**. Once the run has ended, **Response** shows the payload a Response node returned, or `No Response node ran.` **Variables** shows the variables as the run last recorded them. Click a step, or a visited node on the canvas, to open that node's detail. **All Nodes**, a click on an empty part of the canvas, or Escape goes back to the list. The detail shows: - **Visit** — for a node the run reached more than once, which visit to show. - **Result** — the **Exit** the node took, the **Next Node**, and the **Duration**. A node that took its fallback exit shows **Took the fallback exit**. - **Message** — for Send SMS and Send Chat Message nodes, the message the node sent or would have sent. - **Response** — the node's own response data. - **Variables After This Node** — the variables once the node finished, with the names it changed listed under **Changed**. Some values are hidden: - A private variable's value is masked with `*` wherever it appears in the panel, including inside other values. - When **Redact Data in Logging** is on for the workflow or for a node, that node's response, message, and variables show `[REDACTED]`. - When a node's recorded data is over 64 KB, its values show **Too large to display** with the size. ## During a call A voice test shows the call at the top of the **Test Run** panel, above the node list, so you can watch the path on the canvas while you talk. The **Call** bar shows the time elapsed against the five minute limit, and its arrow hides the call controls while keeping **Hang Up** in reach. Turning off the **Run Panel** switch hides the call controls too, but the call keeps going; **Exit Test** hangs it up. - **Keypad** — the digits 0 to 9, `*`, and `#`. A press is sent to the call as a key press, so a Capture DTMF node or a menu reads it as it would from a phone. The digits you've sent show above the keypad. The keys are disabled until the call connects. - **Mute** — stops sending your microphone. - **Hang Up** — ends the call. When the call ends, the panel says why and links to **View call details**: | Message | What happened | | -------------------------------------------------------- | ------------------------------------------------------ | | The workflow ended the call. | The call's channel closed, for example at a disconnect | | Call ended. | You hung up | | Ended after the 5 minute test call limit. | The call hit the time limit | | The call ended at transfer node "…". Nothing was dialed. | The workflow reached a transfer node | The path and the run panel stay on the canvas after the call ends, until you click **Exit Test**. ## Text messages aren't sent A Send SMS node in a test doesn't send a text. It records the message it would have sent (the **To**, **From**, and **Body**) under **Message** in the node's detail, then takes the exit a sent message would. With **Wait for Reply** on, it waits as it would after a real send. This holds for every kind of test, voice test calls included, and for subroutines the workflow calls. The exception is an AI Voice Agent's tools. During a voice test call, both of these **do** send a real text: - A Send SMS node in a workflow the agent runs as a tool. That workflow runs as a normal run. - A Send SMS [node tool](/docs/agents/voice-agents/tools#add-a-node-tool) on the agent. The agent sends it directly, outside any workflow. A Send Chat Message node's message shows under **Message** the same way. ## SMS and chat tests stop at the first wait An SMS or Chat test sends the workflow one message, the **First Message**. When the workflow reaches a node that waits for the next message, such as Send SMS with **Wait for Reply** on, the status shows **Waiting for the next message at** that node. Sending a follow-up message from the editor isn't supported yet. If the waiting node has a timeout, the run takes its no-reply exit when the timeout passes, and the canvas follows it if you are still in the test. ## Transfers dial nothing A test call never places a transfer. - A **Transfer to External** or **Transfer to Queue** node ends the test call when it is reached. The end message names the node, and the call details list the transfer as **Test call ended at transfer** with the node's name and the kind of transfer. The end message waits up to about 30 seconds for that record; if it has not arrived, the panel shows **The workflow ended the call.** instead. - An **AI Voice Agent** node's own transfer tool is refused, and nothing is dialed. The agent is told `This is a test call, so the transfer was not placed.` - A workflow the agent hands off to, and a subroutine the workflow calls, are part of the same test call. A transfer node reached there ends the call the same way. To confirm the transfer destination itself answers, use a real call to a routed number. ## Billing A voice test call is billed as call minutes, the same as a phone call into the workflow. It has no number of its own, so on the Billing page's **Usage** tab its minutes show as **Local call, in**, beside the workflow's **IVR** time. See [How a call is billed](/docs/overview/billing#how-a-call-is-billed). If your organization has no call-minute balance, the call ends before it connects and the panel shows **Couldn't complete the call** with `Your organization's balance is too low to place a test call.` ## Call details A voice test call is recorded as a conversation, and Call Logs lists it as **Browser test call** in place of a caller's number. When it ends, **View call details** opens its page under **Insights → Call Logs**, with the steps the workflow took. See [Call Logs](/docs/workflows/insights-and-reporting#call-logs) for how to read it. ## Limits and defaults | Limit | Value | | ------------------------ | -------------------------------------------------------------- | | Call length | 5 minutes, then the call ends | | Concurrent calls | One at a time per signed-in user, counting agent preview calls | | Version tested | The version open in the editor, as last saved | | Transfers | Never dialed; the call ends at a transfer node | | Messages per SMS or chat | One, the **First Message** | | Records per run | 2,000 records or 8 MB; the step list stops at either | | How long a run is kept | One hour after its last update | | Subroutines | The Subroutine node shows on the path; its nodes don't | The one-call limit is shared with **Call Agent** in the agent editor, so a [preview call](/docs/agents/voice-agents/preview-calls) left running in another tab blocks a test call until it ends or reaches its own five minute limit. ## Troubleshooting **Permission needed.** The browser refused microphone access. Enable it for this site in your browser's settings, then use **Try Again**. **Couldn't complete the call.** The call failed to start. If the message is `Failed to start a test call.`, one cause is another test call or agent preview call still running under your account, for example in a tab you left open. End it, or wait for it to reach the five minute limit, then try again. If the message says the balance is too low, see [Billing](#billing). **Your sign-in session expired.** Reload the page and start the call again. **This test run is no longer available.** The run's record expired an hour after its last update, or the run never started. The path clears from the canvas. Click **Exit Test** and start a new test. **Can't reach the test run. The path may be out of date.** The editor failed to read the run three times in a row. It keeps trying, and the notice clears once a read succeeds. ## Next steps