Where to check what your agent did

Find your question, then go to the screen next to it.

Workbench is where you build. MyCrew is where your agent works for real people. The logs follow the same line, so you always know which screen to open.

Why not put everything in one place? Because customers are not part of building. Keeping their conversations in MyCrew means a test you run while building never mixes with a real customer's chat, and what a customer wrote stays with the people who serve them. Inside MyCrew, though, you do not hop between screens: each message shows its own agent run, so one thread answers the whole question.

If you sell your agent to other businesses: you see nothing of their customers. The business that bought it owns its number and its logs. If they hit a problem, they share that one conversation with you.

Before you connect WhatsApp, the agent should work in Test Run and be released internally. The channel then runs exactly that version. Edits you make afterwards do not reach your customers until you release them, and every release is recorded in the version history.

1. 1
   Open the agent in Workbench and run it in Test Run until it answers as you want.
2. 2
   At the top of the agent page, open Choose internal version and pick the version to go live.
3. 3
   Now open the Channels tab and connect your WhatsApp number. Without a released version, Connect stays disabled and tells you why.

1. 1
   Open MyCrew, then My AI, choose the agent, then Manage.
2. 2
   Open the Channels tab, then WhatsApp.
3. 3
   Under Customer conversations, find the customer. A red Last message not answered badge means the problem is in their latest message.
4. 4
   Click the customer. The message that was not answered has a red box that says why and what to do.
5. 5
   Open Steps in that box to see how far the message got. The last step is where it stopped.

You can only see these cards when you are signed into MyCrew as the organisation that owns the channel. If you belong to more than one organisation, pick the right one at sign-in.

These are the headlines you can see on a message, what each means, and who can fix it.

When the agent did run, a box under the customer's message shows Agent run completed or failed, how long it took, how many BICS it cost and how many steps it had. Open it to see every step with its own time. It shows the shape of the run, never what a step produced and never what the customer typed.

If a message never started the agent, there is no run box. The red reason box tells you why instead, which is the whole point: a message that stops early creates no run, so Runs in Workbench never lists it.

In Workbench, open the agent and its Channels tab. The badge says Connected only when the channel can really reply. If it cannot, the badge says Needs attention and a red card explains why. Below it, Recent problems lists the last few messages that were not answered.

In MyCrew, on the WhatsApp tab, Who can chat has two choices:

• Anyone. Customers just message your number. No account, no linking. You pay for the replies.
• Only people who link their number. For a closed group. Each person connects their number to a BotSpot account first.

Because on "Anyone" you pay for strangers, there are two limits, and you change both. The default is 20 messages per person per hour and 50 BICS per 24 hours for the whole channel. Turn a limit off and it is unlimited, which means anyone who messages your number can spend your BICS without a stop. The card shows what the channel used in the last 24 hours.

The 24 hours are a rolling window, not a calendar day, so the allowance never comes back all at once at midnight. The check is a soft limit. Two messages arriving at the same moment can both pass, so a channel can overshoot by one reply. Your wallet history is the exact record.

If you messaged an AI on WhatsApp and nothing came back, open it in MyCrew, then Channels, WhatsApp, and look at My messages. Each message says Answered or Not answered, how far it got, and what to do. You are never shown the owner's billing or settings. If a message failed, note the short reference code and give it to the AI's owner or to support.

• The reference code on the message that was not answered.
• The time you sent it.
• The agent name and which organisation you were signed in as.

With the reference code, support can read the exact step where the message stopped, so you do not need to describe it.