The AI Dispatch Rules feature automatically hands new, unassigned tickets to an AI agent based on the ticket's department. Unlike the "Run AI agent" action in automatic rules, a dispatch rule needs no trigger and no conditions - you simply declare "for tickets waiting in these departments, run this AI agent and charge this budget", and every ticket that lands in the queue is handed over on its own. It serves as a queue-driven automation layer alongside other LiveAgent AI-powered features.
Note: To use the AI Dispatch Rules feature, it is required to integrate your LiveAgent account with an AI provider, and to have at least one AI agent and one AI budget already configured.
Tip: It is recommended to integrate FlowHunt as your AI provider and set up your AI agents via it to have full control over the workflow and troubleshooting.
FlowHunt, our no-code AI automation platform, is a product of our company, Quality Unit, like LiveAgent. By utilizing our AI features with FlowHunt, we ensure data safety and offer priority support. For customers with active LiveAgent & FlowHunt subscriptions, our support team offers a full initial setup. If you're interested in our free setup service, feel free to contact us at support@liveagent.com.
How it works
A dispatch rule is a small binding of four things: a set of departments the rule watches, the AI agent to run, the AI budget to charge, and a dispatch mode that decides how often the agent may act on the same ticket.
You can create as many dispatch rules as you need. A department may appear in any number of rules - every matching rule fires its own independent run, charged to its own budget. This lets you, for example, run one agent that tags incoming tickets and another that drafts a first reply, both on the same department, each with its own spending limit.
Once a rule is in place, the hand-over happens on its own:
- A ticket enters the to-solve queue without an assigned agent.
- Every dispatch rule that covers the ticket's department queues one dispatch for that ticket.
- A background process picks up due dispatches every 30 seconds and starts an AI agent run for each one, in the same priority order the queue itself uses (by the ticket's SLA due date), so the most urgent tickets are handed over first.
- Once a dispatch fires, it is marked as dispatched and will not fire again for that ticket - unless the rule's dispatch mode allows re-running.
Note: Dispatch rules only apply to unassigned tickets. If an agent takes the ticket before the AI gets to it, the pending dispatch is simply dropped - human ownership always wins over the dispatcher.
When a dispatch is postponed
Before starting a run, LiveAgent verifies that the rule's AI agent and budget still exist, that your AI provider is not currently throttled, and that the budget's monthly limit has not been reached. If any of these checks does not pass, the dispatch is postponed rather than failed - it stays in the queue and is retried on a later cycle.
The retry delay grows with each unsuccessful attempt, starting at one minute and doubling up to roughly an hour. A briefly throttled provider therefore barely delays the ticket, while a rule that stays blocked (e.g. its budget is exhausted) is retried less and less often instead of hammering the provider.
If the rule is deleted, or the ticket is transferred out of the rule's departments, its pending dispatch is discarded instead of fired.
Creating a dispatch rule
To create a dispatch rule, navigate to Configuration > AI > AI Dispatch Rules and click "Create rule". Fill in the following fields:
- AI agent - the agent that will process the tickets covered by this rule.
- Budget - spend for runs started by this rule is charged here, up to the monthly limit.
- Dispatch mode - how often the agent may act on the same ticket. See Dispatch modes below.
- Departments - the departments whose queued tickets the rule watches. At least one department is required; use "Add all" and "Remove all" to select or clear the whole list at once.
Click "Create rule" to save. The rule starts working right away - no publishing or activation step is needed.
Tip: A new rule does not only wait for future tickets. Tickets that are already sitting unassigned in the chosen departments are enqueued for the rule the moment you create it, so the agent starts working through the standing backlog immediately. Very large backlogs are seeded in batches of up to 500 tickets per department; anything beyond that is picked up as those tickets next change state.
Dispatch modes
The dispatch mode controls how many times a single ticket may be handed to the AI agent:
- Once per ticket - the agent acts on each ticket a single time. This is the default and the safest choice.
- Each time the ticket is queued - the agent re-runs whenever the ticket re-enters the queue, e.g. after a customer reply.
What counts as "queued again" is decided by the ticket's lifecycle, not by the queue alone:
- An agent picking up a ticket and putting it back unchanged is treated as nothing having happened - no rule re-fires.
- When a ticket is answered and later returns to the queue on a customer reply, rules set to "Each time the ticket is queued" fire again. Rules set to "Once per ticket" stay silent.
- When a ticket is closed, all of its pending and recorded dispatches are cleared for good.
Tip: Start with Once per ticket. Reserve "Each time the ticket is queued" for agents that genuinely add value on every turn - re-tagging a ticket as its topic evolves, for instance - and keep in mind that it spends credits on every turn of the conversation, not just the first.
Managing an existing rule
All of your dispatch rules are listed under Configuration > AI > AI Dispatch Rules, showing each rule's AI agent, departments, budget and dispatch mode. Click a rule to open its detail view, which is organized into the Rule settings, Departments, Activity and Danger zone sections.
Use the "Change" button in the Rule settings or Departments section to adjust the rule. You can switch its AI agent, budget or dispatch mode, and add or remove departments at any time; the change takes effect from the next dispatch cycle.
To remove a rule, open the Danger zone section and click "Delete dispatch rule". The rule stops applying to new tickets, while tickets it has already dispatched are unaffected - runs that are already in progress finish normally and stay in your reports.
Note: Deleting a dispatch rule cannot be undone. Deleting the AI agent a rule points to does not delete the rule - the rule simply stops dispatching and reports its tickets as retrying until you point it at an existing agent again.
Monitoring a rule's activity
The Activity section of a rule's detail view shows the rule's outstanding work at the moment you open it:
- Pending tickets - tickets that are queued for this rule and still waiting for the agent to act on them.
- Of which retrying - the subset currently held behind a retry delay, typically because the rule's AI agent or budget is missing, or your AI provider is throttled. This line only appears when there is something to report, so a healthy rule shows just the pending count.
A steadily growing "retrying" number is the signal to check the rule's configuration and its budget. Tickets already handed over are no longer counted here - the numbers reflect work remaining, not work already done.
The runs themselves are recorded exactly like any other AI agent run, in the Reports > AI Agent Runs section, where you can review each run's status, the number of tool calls it made and, if it failed, the reason. See the Monitoring AI agent runs article for details.
Credit & budget for dispatch rules
Runs started by a dispatch rule are billed exactly like any other AI agent run: every MCP tool call draws from your account's credit pool, at 1 tool call = 1 credit, charged to the budget you selected on the rule. For a full explanation of how credits and budgets work, please see the separate AI Budgets & credits article.
Because a dispatch rule can fire on every ticket that reaches the queue, it is the highest-volume way to spend credits in LiveAgent. Two habits keep that predictable: grant the AI agent only the tools it actually needs, and set the rule's monthly budget limit deliberately rather than pointing several busy rules at one shared budget.
Reaching a budget limit does not lose tickets. When the monthly limit is hit, the rule stops firing and its tickets simply accumulate as pending (and retrying). They are dispatched once the limit resets or you raise it, so no ticket is silently skipped.
Example use cases
First-response triage on a general inbox
Point a dispatch rule at your general support department with an AI agent that has the "Get ticket messages", "Add tags" and "Transfer ticket" tools enabled, and leave the dispatch mode at Once per ticket. Every ticket arriving in the shared inbox is read, tagged by topic and moved to the right specialist department within moments - before any agent has to open it. Unlike an automatic rule, you do not need to define a trigger or maintain conditions; anything that reaches the queue is covered.
Resolving repetitive requests automatically
For a department that receives a high volume of a few predictable questions - order status, password resets, opening hours - create a rule with an agent allowed to answer and resolve tickets. The agent handles what it can as soon as the ticket is queued, and anything it cannot resolve stays in the queue for your team, so agents only see the tickets that actually need them.
Out-of-hours coverage
Dispatch rules act on tickets as they are queued, regardless of the time of day, which makes them a natural fit for nights and weekends. An agent that acknowledges the request, gathers the missing details and tags the ticket keeps customers from waiting until morning for a first reply, and leaves your team a fully prepared ticket at the start of the shift.
Re-tagging long conversations
When a conversation's topic tends to drift - a billing question turning into a technical one - use the Each time the ticket is queued mode with a tagging agent. Each time the customer replies and the ticket returns to the queue, the agent re-reads the conversation and updates the tags, so your reporting and routing reflect what the ticket is about now rather than what it was about when it arrived.