Working the Tickets Board

Where the Board Lives

The built-in ProxyAI desk gives you a board of every ticket the bot has opened. It appears once Helpdesk Connect is set to the built-in desk — with Gorgias, Zendesk, Freshdesk or Help Scout connected instead, you work those tickets in that desk and this board stays empty.

• Web dashboard — My Products → Support Tickets on the bot's card, or the Support Tickets link inside the Helpdesk tab.
• Shopify app and WordPress plugin — the Tickets tab.
• Wix app — the Tickets tab in the ProxyAI dashboard inside your Wix site.

The Columns

• Open — needs you.
• Staged — the Smart Helpdesk Agent has written a reply and is waiting for you to approve it. Empty unless you have the agent switched on.
• Pending — you replied; waiting on the customer.
• Resolved — done.

Sending a reply moves a ticket to Pending. The customer writing back moves it to Open. Resolving hands the conversation back to the bot.

You can drag cards between columns. The search box looks at ticket number, name, email, subject and category. Current and Archived are separate views, and anything archived can be brought back.

Working a Ticket

Open a card to read the whole conversation — the customer's messages, the bot's answers, and any pictures they sent. Scroll the thread; long conversations start at the top.

Reply in the box at the bottom. Your answer goes out on the chat channel the customer was using, and by email too if you set up a sender address in the Helpdesk wizard.

If the bot said it opened a ticket and none appears here, check a desk is still connected — the bot can only file tickets while one is.

The Smart Helpdesk Agent

The agent reads each new ticket and writes the reply for you. It does not send anything by default — the draft waits in the Staged column until you approve it. You see what it would have said before a customer ever does.

Find it in Tickets → Agent mode.

Switching it on

1. Enable the agent on this desk. Off by default, per desk. Turning it off takes effect on the next ticket — no waiting.
2. Model. Choose from the models we allow for ticket work. This is deliberately a shorter list than the chat bot's: a drafted refund reply that invents policy is worse than no draft at all. If your chosen model is unavailable when a ticket arrives, we use our default for you and the board tells you which model actually wrote it.
3. Instructions. Your voice and your policy, in plain words — *"refunds under $20 approve directly; never promise delivery dates; sign as Sarah."* These sit alongside the built-in safety rules; they do not replace them.
4. Caps. Drafts per ticket (3 by default) and drafts per day (100). The first stops the agent going back and forth with one unhappy customer; the second is your spend ceiling.

Reviewing a draft

A staged card shows the draft with the model that wrote it. You can:

• Approve and send — it goes to the customer through exactly the same path as a reply you typed yourself.
• Edit, then send — change anything first; your edit is what the customer receives.
• Dismiss — nothing is sent. The draft is kept in the ticket's history.

When the agent will not answer

Sometimes there is nothing to answer from. If the ticket needs a policy you have not written down, or details only a person can find, the agent says so instead of guessing: the ticket stays in Open with a short note explaining what it would need. That note is for you, never for the customer — no padded reply is ever sent in place of a real answer.

The fix is usually to give it the knowledge, not to switch it off.

Giving the agent its own knowledge

The agent has a Knowledge section of its own, with two sources you control separately:

• Search bot knowledge — reuse the documents your chat bot already answers from. Each search costs credits at the Bot Knowledge rate.
• Internal documents — a private library only the agent reads. Your chat bot never sees these. It is the right place for the things you would tell a new support hire and not a customer: approval limits, when to escalate, what staff may promise, your refund SOP.

Upload the same file types Bot Knowledge accepts. A short, plainly written policy document does more for draft quality than a longer model does.

YOLO mode — letting it send on its own

YOLO mode lets the agent send confident replies directly, without waiting for you. Anything it is unsure about still lands in Staged.

It is locked until you have approved 10 of its drafts. That is deliberate: you should see what the agent would have sent, ten times, before you let it send. The counter is on the toggle.

Even with YOLO on, the daily cap still applies, the per-ticket cap still applies, and a ticket the agent cannot answer still stops in Open for you.

Every reply the agent is about to send on its own is also checked by a leak screen first. It compares the draft with what the customer actually asked, with your instructions and the documents the agent used in view — the internal knowledge base included. A reply that gives the customer the refund window for their order goes out, even when that fact came from your internal notes; a reply that pastes those notes, lists your documents, or hands over a whole policy because someone asked to "see the full document" stops in Staged with the reason on the card. If no model is available for the screen on the platform it cannot run, and the agent stages everything until it can.

Replies the agent sends on its own reach the customer on their chat channel like any other reply. Whether you tell your customers that a reply was AI-assisted is your own policy — the tools are here either way.

What it costs

Drafting is metered like any other AI usage, on the ticket_agent line in Usage & costs, from the same balance. In practice a draft costs a fraction of a cent; searching your knowledge for grounding usually costs more than the writing does. Set the daily cap to whatever ceiling you are comfortable with.

One thing to know before you enable it

While the agent is on, the plain "we got your ticket" acknowledgement email is not sent — the agent's reply is meant to be the first thing the customer hears. If you have switched the agent on and customers seem to hear nothing, check the board: tickets sitting in Open with a "needs a human" note are waiting on you, not on the agent.

Test It

1. Open a ticket the way a customer would — through the chat, or your ticket form.
2. Watch the board: the ticket appears in Open, and within a minute or so a draft appears in Staged if the agent is on.
3. Read the draft. Approve it, and confirm the answer arrives in the original chat.
4. Try one the agent cannot know — something your knowledge does not cover — and confirm it stays in Open with a note rather than inventing an answer.