How to Use LiveAgent to Triage Tickets With an AI Agent

The short answer: if every ticket lands in one shared inbox and someone sorts it by hand each morning, one LiveAgent AI agent can tag each new ticket and move it to the right department, and the Starter plan at $19 per agent a month includes exactly one AI agent, with no AI fee from LiveAgent (monthly billing, checked on 26 October 2026 on the pricing page). The steps: connect an AI provider, write down your categories, create the agent with four tools, give it a budget, point a dispatch rule at the inbox, and have a person check its work daily for the first week.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

Before you build it:

1. Count the tool calls. A triage run reads the ticket, lists the departments, adds a tag and transfers the ticket: four tool calls. For example, 1,500 tickets a month x 4 = 6,000 tool calls, against the Default budget of 300,000 a month that LiveAgent creates for every account (AI budgets). The token bill goes to your AI provider, not to LiveAgent.
2. Take the categories your team already uses. Pull the last 50 tickets and write down, for each, the tag a person gave it and the department it went to. Those pairs are the agent's instructions; there is nothing new to invent.
3. Fix the check before launch. Every run is logged in Reports > AI Agent Runs with its status, tool calls and failure reason (run monitoring). Plan to compare 20 of the agent's tickets a day with the tag a person would have given.

What you need

• A LiveAgent account. The 30-day trial needs no card and lets you test every feature, per the FAQ on the homepage.
• An API key from OpenAI, Anthropic or Google Gemini, or a FlowHunt account; those are the four providers LiveAgent lists (AI providers).
• The departments and tags from your list, created before the agent runs. It can only move tickets to departments that exist.

Step 1: Connect the AI provider

In the OpenAI dashboard, open API keys, create a secret key and copy it straight away, because OpenAI shows it only once. In LiveAgent, go to Configuration > AI > Setup AI provider, click "Add AI provider", choose OpenAI, paste the key and name the integration (OpenAI integration). Anthropic and Gemini follow the same pattern from their own consoles.

Step 2: Create the agent and write its instructions

Go to Configuration > AI > AI Agents and click "Create AI Agent". Name it after its job, such as "Inbox triage": the name shows up in rule actions, in the run report and in the system message logged on each ticket. Select your provider and model, then write the instructions, which act as the system prompt and can run to 5,000 characters (OpenAI setup guide).

Write them from your list of 50: one line per category with its tag and its department, a sentence on how to tell close categories apart, and a fallback for anything that fits none of them, such as "tag needs-review and leave the ticket where it is".

Step 3: Allow four tools, and no more

LiveAgent's use-case article builds tagging and routing from "Get ticket messages", "Add tags", "List departments" and "Transfer ticket". Tick those and leave "Resolve ticket" and "Assign ticket" off. A tool you do not allow is unavailable to the provider even when the model asks for it (AI agents overview), so the tool list, not the prompt, is the guardrail.

Step 4: Give it a budget of its own

Create a budget for triage under AI budgets instead of drawing on Default. A budget caps the tool calls its agents can make in a calendar month and resets on the first; the list highlights it at 80% and again at the limit, and once it is used up the runs fail until the next month (AI budgets). For example, 6,000 expected calls plus half again for busy weeks is 6,000 x 1.5 = 9,000.

Step 5: Point a dispatch rule at the inbox

Go to Configuration > AI > AI Dispatch Rules, click "Create rule" and choose the agent, the budget, the dispatch mode and the departments to watch. "Once per ticket" suits triage; "Each time the ticket is queued" re-tags a ticket whenever the customer replies, at the cost of more tool calls. A background process picks up waiting tickets every 30 seconds, most urgent first, and only unassigned tickets are handed to the agent (dispatch rules).

If you need conditions, such as only tickets from one mailbox, use an automatic rule instead: Configuration > Automation > Rules, trigger "Ticket created", action "Run AI Agent", with the same agent and budget.

Step 6: Have a person check the agent

The agent's action is logged on each ticket as a system message, so the check happens where your team already works. For the first week, one person reviews the 20 sampled tickets each day, corrects wrong tags and notes the pattern; when the same miss appears twice, add a line to the instructions. In the run report, Completed means the agent made a closing call such as Transfer ticket, and a Failed run carries its reason, from "AI provider error" to "Tool execution failed" when a department or tag it names no longer exists (run monitoring).

This agent never writes to a customer. Article 50 of the EU AI Act, which applies from 2 August 2026, asks that people are told when they interact with an AI system, at the latest at the first interaction (Regulation (EU) 2024/1689 on EUR-Lex, Articles 50 and 113). A tagging agent does not interact with anyone; the chatbot on your chat button does, so add the notice to its greeting on the day you switch it on.

Warnings in LiveAgent's own articles

• The backlog goes first. A new dispatch rule starts on the tickets already in the chosen queues, in batches of up to 500 per department (dispatch rules). For example, a backlog of 1,200 tickets x 4 calls is 4,800 tool calls in the first hours, so create the rule when the queue is short, or try it on one small department.
• Every extra tool is extra risk. The overview advises granting only the tools an agent needs, which also keeps budget use down (AI agents overview).
• An empty budget stops the line. When a dispatch rule's budget is spent, tickets pile up as pending and retrying until the limit resets or you raise it (dispatch rules).
• A throttled provider pauses runs. They resume on their own, and a run that cannot resume within 60 minutes times out (run monitoring).

Choosing the provider

LiveAgent's AI-agent setup articles recommend FlowHunt as the provider, and the pricing page describes FlowHunt as LiveAgent's own AI platform: the default advice keeps your AI spend with the same company that sells you the helpdesk. OpenAI, Anthropic and Gemini connect with an API key and run the same agents, so pick the provider whose contract and data terms your company has already approved.

What a week on the trial settles

Steps 1 to 6 fit inside the free trial, which covers every feature according to the homepage FAQ. Seven days of triage on real tickets, with the daily sample, tell you whether the agent sorts as well as the person who did it by hand; if it does, the paid plan buys the same result for longer, not a better agent. If it does not, the run log shows where: wrong categories in the instructions, a missing department, or a model that needs a larger brief.

FAQ

Can the agent answer customers too?

Yes, if you allow it the tools: LiveAgent's use-case article describes an agent that answers and resolves a few predictable requests such as order status. That makes it customer-facing, so the AI disclosure applies, and the Resolve ticket tool deserves its own budget and its own sample check.

How many AI agents does each plan include?

One on Starter, 5 on Team, 50 on Business and unlimited on Enterprise; one agent working in several places still counts once (pricing page).

Do I need FlowHunt?

No. FlowHunt is one of four providers, built by the same company; OpenAI, Anthropic and Gemini work too (AI providers). The glossary entry covers the plans and what each includes.