AI Work feature overview

Article navigation:

• Overview
• How AI Work Relates to Other AI Features
• Prerequisites
• Getting Started
  • Step 1 — Create an AI work definition
  • Step 2 — Create MCP token and connect it in your flow
  • Step 3 — Trigger AI work from an automated rule
• MCP Server and authorization Tokens
• Allowed Tools Reference
  • Read tools
  • Write tools
• Monitoring Executions
  • Execution statuses
  • Failure reasons
• Tool Call Budgets
• Example Use Cases
• Advantages Over Manual FlowHunt Rule Actions
• Frequently Asked Questions

Overview

The AI work feature lets you define reusable AI-powered workflows and trigger them automatically on tickets through LiveAgent's automated rules engine. Instead of manually configuring an HTTP call inside every rule, you set up a named work definition once — choosing which AI provider to use, which flow to invoke, and which actions the AI is permitted to take — and then reference that definition from as many rules as you like.

When a rule fires, LiveAgent dispatches the work definition to your AI provider (currently FlowHunt), which processes the ticket and performs actions such as tagging it, resolving it, assigning it to an agent, or transferring it to another department — all without any human intervention.

How AI Work Relates to Other AI Features

LiveAgent provides several AI-powered capabilities. AI work sits alongside them as a background automation layer:

Prerequisites

1. FlowHunt account — You need a FlowHunt workspace with at least one AI flow (agent) configured to process LiveAgent tickets.
2. FlowHunt AI provider integration — Go to Configuration > AI > Setup AI provider and connect your FlowHunt API key. See Integrating FlowHunt as an AI provider for step-by-step instructions.
3. AI agents — You need to create some AI agents under Configuration > AI > AI Agents who will execute your AI Work.

Getting Started

Step 1 — Create an AI work definition

A work definition is the central configuration object. It tells LiveAgent which AI provider to call, which specific flow (agent) to invoke, and which actions the AI is allowed to perform on the ticket. You can create as many definitions as you need — one per use case is a common pattern (e.g., one for auto-tagging, one for auto-resolving common FAQ tickets).

1. Navigate to Configuration > AI > AI work tab and click Create AI work.
2. Enter a descriptive Name (e.g., Auto-classify inbound email). The name appears in rule actions and in the execution log, so choose something that clearly identifies the workflow's purpose.
3. Under AI provider, select the FlowHunt integration you set up earlier.
4. Under Flow, choose which FlowHunt agent (flow) should be invoked. The dropdown lists all flows available in the connected workspace.
5. Under Allowed tools, select which actions the AI agent is permitted to take on the ticket. See the Allowed Tools reference below for a full list.
6. Click Save.

Step 2 — Create MCP token and connect it in your flow

1. Edit the flow you wish to use in FlowHunt, click your AI Agent and under Tools click Add Tool.
2. Search for and select the MCP Client tool.

3. Click the Edit Servers button and toggle the Advanced Mode and copy the following into the MCP configuration text area:

   {
     "External server": {
       "url": "https://YOUR_ACCOUNT.ladesk.com/public/api/mcp",
       "transport": "sse",
       "headers": {
         "Authorization": "Bearer YOUR_TOKEN"
       }
     }
   }

4. Switch back to LiveAgent and edit the AI work you've created in Step 1.
5. Under the MCP Tokens section click the Generate Token button and copy the Token. and replace YOUR_TOKEN in FlowHunt's MCP configuration with it.
6. Replace YOUR_ACCOUNT in the FlowHunt's MCP configuration with your LiveAgent account's URL or get the whole MCP URL from the Edit AI work screen.

Step 3 — Trigger AI work from an automated rule

1. Navigate to Configuration > Automation > Rules (or Time Rules for time-based triggers) and click Create.
2. Set a Trigger for the rule — for example, Ticket is created, Message is received, or Tag is added to ticket.
3. Add any Conditions you need (e.g., only tickets in a specific department, or tickets with a particular subject keyword).
4. In the Actions section, click Add action and select Start AI work.
5. In the action parameters:
   • AI work — Choose the definition you created in Step 1.
   • Agent — Select to which AI agent should the actions performed by the AI (e.g., adding a tag, resolving a ticket) be attributed in the ticket history.
6. Click Save rule.

From this point forward, every time the rule fires, LiveAgent will dispatch the selected AI work definition against the matched ticket. The execution runs in the background — no further manual steps are needed.

MCP Server and authorization Tokens

LiveAgent exposes a built-in MCP server (Model Context Protocol) that FlowHunt connects to when executing an AI work flow. When the FlowHunt flow needs to read ticket data or perform an action, it calls LiveAgent's MCP server endpoint and LiveAgent executes the corresponding tool. This is the communication channel that makes the allowed-tools permission list enforceable — every call is authenticated and validated before it reaches the ticket.

The MCP server endpoint is: https://<your-liveagent-domain>/public/api/mcp and the actual URL for your LiveAgent account is displayed here.

Tokens are valid for two years so you might need to rotate them if you keep using given AI work / FlowHunt flow for more than 2 years.

All tokens for an AI work definition are revoked automatically when the definition is deleted.

Allowed Tools Reference

Each work definition includes a list of allowed tools — the specific read and write actions the AI agent may invoke when processing a ticket. Tools not in the allowed list are unavailable to the AI for that definition, even if the underlying FlowHunt flow would otherwise request them.

Read tools

Write tools

Monitoring Executions

Every time AI work is triggered for a ticket, LiveAgent creates an execution record that tracks its full lifecycle. To review executions, navigate to Reports > AI Work Executions.

Each execution record shows:

• The AI work definition that was executed
• The ticket that was processed
• The AI agent which executed the work
• Start time and duration
• Number of tool calls made
• Current status and, if applicable, the failure reason

Execution statuses

Failure reasons

Tool Call Budgets

To protect system stability and prevent runaway AI processes, LiveAgent enforces a shared tool call budget across all AI work executions in your account. Every read or write action the AI performs on a ticket counts as one tool call and draws from this shared pool. Each AI work execution runs under the AI agent you selected in the rule action. Its tool calls count against the AI agent seat limits.

The actual budget limits are determined by your subscription plan. By default each AI agent can make up to 1,000 tool calls in an hour and 10,000 tool calls in a day.

You can view your current available total budget under Reports > AI Work Executions. This budget scales automatically with the number of active human and AI agent seats in your account, so larger teams automatically have a proportionally larger shared pool. Budgets reset at the start of each hour and each calendar day.

Example Use Cases

Automatic ticket classification and tagging

Create a work definition with the Get ticket messages and Add tags tools enabled. Build a FlowHunt flow that reads the first message, identifies the topic (billing, technical issue, feature request, etc.), and applies the corresponding tag. Attach this definition to a rule that triggers on Ticket is created for your general-support inbox. Every inbound ticket will be tagged within seconds of arrival, ready for agents to sort and prioritize.

Auto-resolve common FAQ tickets

Enable Get ticket messages, Add tags, and Resolve ticket for a definition. Configure the FlowHunt flow to check whether the ticket contains a standard question that can be handled without agent involvement, apply a tag such as auto-resolved, and then resolve the ticket. Use rule conditions to restrict this to specific departments or sender types so the AI only auto-resolves tickets where that is appropriate.

Intelligent ticket routing

Enable Get ticket messages, List departments, and Transfer ticket. Configure the flow to read the ticket content, determine the correct specialist team, and transfer the ticket automatically. This is especially useful when tickets arrive through a generic inbox and need to reach the right department before an agent picks them up.

SLA-aware escalation with Time Rules

Use a Time Rule instead of an instant rule. Set the trigger to fire when a ticket has been open for a defined period without a response. The AI work definition can read the ticket context, assign it to a senior agent, and add an escalation tag — all without a supervisor having to manually monitor the queue.

Advantages Over Manual FlowHunt Rule Actions

LiveAgent previously supported triggering FlowHunt flows from rules using the generic HTTP request action (see Triggering FlowHunt AI agents via rules). The AI work feature is the recommended replacement for that approach and offers several important improvements:

Frequently Asked Questions

Can I use AI work alongside other actions in the same rule?

Yes. The Start AI work action can be combined with other rule actions in the same rule. For example, you could send an auto-reply email and trigger AI work in the same rule, so the customer receives an instant acknowledgement while the AI processes the ticket in the background.

What happens if the same ticket matches two rules that both trigger AI work?

Each matching rule creates a separate, independent execution. Both executions run in parallel. If both try to write to the ticket (for example, both try to resolve it), the second write will apply after the first completes. To avoid conflicts, use rule conditions carefully to ensure that overlapping rules target different ticket states or departments.

Can I trigger AI work manually, outside of rules?

AI work is designed primarily for rule-based automation. Ad-hoc triggering from outside a rule is not currently supported in the LiveAgent panel, however you could create a rule which executes AI work when a specific tag is added to a ticket and then just tag the tickets on which you want your AI work to execute.

Does AI work appear in the ticket's conversation history?

Yes. If the AI performs write actions (adding tags, resolving a ticket, assigning it, transferring it, etc.), those actions are recorded in the ticket's activity history and attributed to the AI agent identity you selected when configuring the rule action. Agents can see exactly what the AI did and when.

Can I stop or cancel a running execution?

Executions that are actively Running cannot be cancelled mid-flight from the panel. Executions in the Waiting state will resume automatically once the provider's throttle window expires. If an execution appears to be running for an unexpectedly long time, check the FlowHunt flow for loops or misconfigured steps that may be causing excessive tool call usage.

Is my ticket data sent to third-party AI providers?

The ticket content (messages and metadata) is transmitted to your configured AI provider — FlowHunt — for processing. FlowHunt is developed by Quality Unit, the same company behind LiveAgent, and adheres to the same data safety and privacy standards. If your FlowHunt flow uses an external language model (e.g., OpenAI GPT-4 or Anthropic Claude), your data will also be subject to that provider's terms. Review your FlowHunt flow configuration to confirm which underlying models are in use.