> For the complete documentation index, see [llms.txt](https://en.help.firstline.cc/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://en.help.firstline.cc/feature/ai-agent.md). # AI Agent Learn what FIRST LINE's Fancy AI agent can do, and how to configure reply instructions, knowledge, and handoff flows. {% hint style="info" %} Fancy AI agent is expected to be available in **Q4 2026** for use. The actual launch time and feature details are subject to FIRST LINE's official announcement. {% endhint %} ## Meet Fancy {% columns %} {% column width="16.666666666666664%" %} {% endcolumn %} {% column width="83.33333333333334%" %} **Fancy is FIRST LINE's AI agent, powered by an advanced large language model and designed to understand customer needs, hold multi-turn conversations, and help handle everyday customer service tasks.** Fancy can determine the appropriate response based on the customer's question, retrieve available knowledge content, and, when needed, hand the conversation over to a human agent or a designated conversation script, allowing automated replies to work together with existing service processes. {% endcolumn %} {% endcolumns %} Fancy is FIRST LINE's **AI agent**It can understand customer questions, continue responding based on the conversation context, and provide assistance according to the company's configured approach. When a task requires fixed steps, Fancy can hand the conversation over to a conversation script; when human assistance is needed, it can also transfer it to a human agent. For example, if a customer first asks, “Can I return the product?”, and then follows up with “Can I return it after opening the package?”, Fancy can continue the previous topic and answer based on the information provided by the company, without requiring the customer to explain everything again each time. {% hint style="info" %} Fancy's replies are influenced by the information and settings provided by the company, and incomplete or incorrect content may still occur. Important pricing, refund, and service commitments should be based on the company's actual policies. {% endhint %} ## What tasks can Fancy help with? | Feature | How it can be used | | ------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------ | | **Understand questions, maintain conversations** | Answer customers' natural-language questions and continue based on the context. | | **Query company knowledge** | Refer to relevant information in the knowledge base, products, and AI training resources. | | **Follow service instructions** | Set the brand tone, response style, and principles for handling special cases. | | **Hand off fixed processes** | When a customer needs to check an order or submit a request, transfer them to a prebuilt conversation script. | | **Transfer to a human agent** | Depending on the customer's needs or the company's settings, hand conversations that require human assistance to a service representative. | ## Get started with Fancy {% stepper %} {% step %} ### Create an AI agent Log in to FIRST LINE and open the **AI agent**management page, add Fancy, and fill in the name and description. {% endstep %} {% step %} ### Set basic information Choose a conversation avatar, and set idle time, keywords for transferring to a human, the number of consecutive failed answers, and the maximum number of conversation turns as needed. {% endstep %} {% step %} ### Write response instructions Go to **Instructions**and set the principles that all conversations should follow, such as the tone of replies, which information needs to be confirmed, and which matters cannot be promised on its own. For special cases, you can add scenario-based instructions separately. {% endstep %} {% step %} ### Choose the knowledge scope Go to **Knowledge**and select the knowledge sources that Fancy should refer to. You can choose knowledge bases, products, and AI training resources as needed, and use the filters on the screen to narrow the scope. {% endstep %} {% step %} ### Set cases that need to be transferred Go to **Routing**to define which customer needs should be transferred to a human agent or handled by a designated conversation script. For example, when a customer wants to formally submit a return request, switch to the return request script. {% endstep %} {% step %} ### Try it out and enable First use**trial chat**to confirm common questions, follow-up questions, and transfer scenarios. After confirming that the replies meet service requirements, enable Fancy and select this AI agent in the message channels where it will be used. {% endstep %} {% endstepper %} ## Common settings guide ### Basic settings and transfer to a human | Setting | Description | | ------------------------------------ | --------------------------------------------------------------------------------------------------------------------------- | | Name, description | Helps identify the purpose of different AI agents. | | Conversation avatar | Determines the image Fancy displays in conversations. | | Enable / disable | Controls whether the agent provides service. | | Idle timeout | Ends the conversation according to the settings when the customer has not interacted for a long time. | | Keywords for transfer to human | When customers mention specified words, the conversation is handed to a human agent according to the rules. | | Number of consecutive failed answers | When no suitable answer can be found multiple times, the conversation is handed to a service representative for assistance. | | Maximum number of conversation turns | After the conversation reaches the set number of turns, it is transferred to a human according to the rules. | ### Instructions: deciding how Fancy should respond **Default instructions**are the principles that should be followed in every conversation;**scenario-based instructions**are used to supplement how special cases should be handled. For example: > You are the store's online customer service. Please respond to customers in Traditional Chinese, clearly and politely. When dealing with return or exchange issues, prioritize the service policies provided; if the information is insufficient, first confirm the necessary details and do not promise refund results on your own. It is recommended to keep product information, pricing, and service policies in**Knowledge**maintenance, so the instructions can focus on how to respond. For detailed examples, refer to [How do you write AI agent instructions correctly?](/feature/ai-agent/how-do-you-write-ai-agent-instructions-correctly.md). ### Knowledge: information needed for responses In the **Knowledge** tab, choose the content Fancy can refer to, such as FAQs, product data, or AI training resources. If you only want to use specific categories or content, you can filter using the criteria provided on the screen. **Please update the knowledge content regularly.** If the data is incomplete, Fancy may not be able to give a definite answer. ### Routing: deciding when to hand off In the **Routing** Set scenarios and handling methods in the tab, such as: * When the customer explicitly submits a return request,**switch to a conversation script**to collect information through fixed steps. * When the customer asks for a human agent or encounters an issue that requires human handling,**Transfer to a human agent**. Fancy will determine whether it meets the configured conditions based on the conversation content. It is recommended to use clear and specific wording, and to verify the results through trial chat. ## When should it be used together with a conversation script? Fancy is suitable for answering more open-ended questions; conversation scripts are suitable for situations that require fixed menus, data collection, or clear handling steps. The two can be used together: **Customer asks → Fancy understands the need → Conversation script executes the fixed process → End or continue service according to settings.** Please note:**After a conversation script finishes normally, Fancy will not directly answer again.** If you want Fancy to continue handling the conversation within the same thread, use the **“Return to AI agent”** function in the conversation script; please test the full flow through trial chat before正式 use. Further reading:[Learn about handing off from an AI agent to a conversation script](/feature/ai-agent/learn-about-ai-agent-handoff-conversation-scripts.md)、[How do Fancy and conversation scripts work together?](/feature/ai-agent/start-ai-agent-from-scenario.md). ## Recommendations before use * First confirm which issues Fancy should handle and which matters should be handed to a human agent. * Organize the latest product information and service policies into the knowledge sources. * Keep the instructions clear and concise, and avoid conflicting requirements. * Before正式 use, test situations such as general questions, insufficient information, follow-up questions, handoff, and conversation ending. * Refunds, fees, contracts, and other important matters should still be handled according to the company's approved service process.