> For the complete documentation index, see [llms.txt](https://help.pickyassist.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.pickyassist.com/setting-up-guide/ai-agent/getting-started.md). # Getting Started Get Started with Ai Agents ### Add AI Agent to a flow 1. Open[ **Flow Builder**.](https://app.pickyassist.com/flows?flow) 2. Open an existing chatbot flow or create a new one. 3. Drag **AI Agent** onto the canvas. You can also right-click the canvas and select **AI Agents**. 4. Connect the AI Agent to the required flow steps. 5. Select the block and choose **Configure Agent**.
### Configure instructions The **Instructions** tab defines the agent's behaviour. Select an AI persona, then define:
* What the agent should do. * What the agent should never do. * When the conversation should be marked as completed.
#### Select an AI persona Choose a persona that matches your use case. Common options include: * Sales Assistant * FAQ / Support * Lead Qualification
Selecting a persona gives you a starting instruction template. You can customize it for your business. {% hint style="info" %} Changing a customized agent's persona does not overwrite its instructions automatically. Choose whether to use the new template or keep your existing instructions. {% endhint %} #### Instruction fields | Field | Purpose | | ------------------------------------------------------- | -------------------------------------------------------------------------------------------- | | What should it do? | Define the agent's tasks, greeting, questions, responses, and conversation handling. | | What should it never do? | Define restrictions, such as avoiding promises, guesses, and unsupported information. | | When should the AI mark this conversation as completed? | Define when the conversation reaches the **Completed** path and moves to the next flow step. | ### Build the Knowledge Base The **Knowledge** tab contains information the AI Agent can use in customer responses.
You can provide information through: 1. **Add Information** 2. **Upload Documents** 3. **Products & Services**\ \
**Add Information** Add short, text-based topics that the AI Agent can reference. For best results: * Keep each topic focused on one subject. * Keep topics below approximately 2,500 characters. * Split long content and large tables into separate topics. * Avoid duplicate or conflicting information.
For example, use separate topics for **Course Details**, **Course Fees**, **Batch Timings**, **Admission Process**, and **Refund Policy**. **Upload Documents** Upload supported documents when the agent needs large reference files. Supported formats include `PDF`, `DOC`, `DOCX`, `XLS`, `XLSX`, `PPT`, `PPTX`, and `TXT`.
You can upload up to two PDFs per agent. Each document can be up to 15 MB. {% hint style="warning" %} Document processing can cost up to 10× more per response. Use **Add Information** for smaller content. Keep uploaded documents accurate and current. {% endhint %} **Products & Services** Products and services let the AI Agent present your offerings in conversations.
Add catalog entries in either mode: * **Manual** — Enter product or service details directly. * **Dynamic** — Fetch details from a configured data source. Both modes create entries the agent can present and reference. {% hint style="info" %} Add only information the agent should use. If information is unavailable, instruct it to decline politely or hand off the conversation. {% endhint %} **Add dynamic Knowledge Base or product data** Use Attributes to pass dynamic knowledge and product details to the AI Agent. For example, pass a previous AI conversation summary as context. You can also pass Picky Assist CRM fields, such as requirements and notes. To fetch CRM data, add a **Data Lookup** step before the AI Agent. Use the **Search** option, then pass the result through Attributes. To use third-party data, add an **HTTP API** block before the AI Agent. Pass the API response to the Knowledge Base through Attributes. {% hint style="warning" %} The built-in Test Widget may not support dynamic data. Publish the flow and test the live flow to verify dynamic responses. {% endhint %} ### Collect and extract data The **Collect & Extract Data** tab lets the agent collect customer details and extract data from customer content.
**Data fields to collect** Data fields are details that the AI Agent actively asks the customer to provide.\ Click **Add a Field** to add the necessary fields for collecting information from customers.
Examples include: * Name,Email,Mobile,Date. When adding a field, select the appropriate field type based on the information you want to collect from the customer. **Text:** Collects text-based information such as name, address, or preferences. **Number:** Collects numerical values such as age, quantity, or budget. **Phone:** Collects the customer's phone number. **Image:** Allows customers to upload an image. **Email:** Collects the customer's email address. **Date:** Allows customers to provide or select a date. **Buttons:** Provides predefined options that customers can select with a tap. **Document:** Allows customers to upload a supported document or file. Configure up to 15 fields per AI Agent. Mark a field as **Required** only when the agent must collect it before completing the conversation. {% hint style="info" %} Use required fields sparingly. Keep questions to three or four when possible, or customers may stop responding. {% endhint %} **Extract data** Extract data fields let the AI Agent identify information from messages, documents, and images automatically.
For example, an uploaded certificate can provide: * Name * Qualification * Year of completion * Institution The agent stores extracted values in their matching Attributes. Use them to update Picky Assist CRM or send values to an external CRM through the [**HTTP API**](/setting-up-guide/setting-up-flow-builder/steps/trigger-http-api.md) block. Unlike Collect fields, Extract fields do not need a separate question for each value. {% hint style="warning" %} Configure extraction fields that match the information customers are expected to provide. {% endhint %} **Use Attributes in later steps** Attributes store data collected or extracted by the AI Agent. This includes customer answers and data from documents or images.
The configured Attribute name becomes available in later flow steps. Use it to update Picky Assist CRM, define conditions, or personalize messages. #### **Pushing Collected or Extracted Data to 3rd-Party Platforms** To send collected or extracted information to third-party platforms such as CRM or ERP systems, use the [Trigger HTTP API](/setting-up-guide/setting-up-flow-builder/steps/trigger-http-api.md) step in the flow and map the relevant Attributes to the corresponding fields in the API request.
### Configure handoff The **Handoff** tab transfers a conversation to a human agent or another AI Agent when the specified conditions are met.
**To add a Handoff path** * Enter a **Path Name** to identify the purpose of the handoff, such as humen handoff or sales team . * **Trigger this when**, specify the condition that should trigger the handoff. \ For example, when the customer requests to speak with a humen agent You can configure up to 10 handoff paths. A path can trigger when a customer message contains a configured keyword or phrase. Connect each handoff output to its required next flow step. {% hint style="warning" %} Test the complete handoff flow after publishing. Confirm that every configured action runs correctly. {% endhint %} {% hint style="success" %} You can hand off to another AI Agent. Specialized Sales, Support, and Billing agents can share the conversation context. {% endhint %} ### Configure follow-up The **Follow-up** feature sends messages when a customer stops responding.
You can configure a **follow-up** by specifying **Send After :** Specify how long the AI Agent should wait before sending the follow-up message. **Instructions :** Provide instructions for the AI Agent on what to include in the follow-up message and how it should respond based on the conversation. You can configure up to three follow-up messages. {% hint style="info" %} Keep follow-ups short, relevant, and helpful. Avoid repeated messages sent too frequently. Follow-up works only during the 24-hour free-form session and suits sales-focused agents. {% endhint %} ### Configure settings The **Settings** tab controls the AI provider, model, file and image processing, reasoning, conversation limits, and related behavior.
Configure how your AI Agent responds, processes customer inputs, and manages conversations using the following settings. **AI Provider**\ Select the AI provider, such as ChatGPT or Gemini.
**Enable Reasoning:** Enables additional reasoning for supported models. This can improve the handling of complex decisions but may increase response time and cost. {% hint style="info" %} Enable reasoning only for complex, multi-step decisions. Standard models suit straightforward questions and data collection. {% endhint %} **Model**\ Select a model from the chosen AI provider. Recommended models are GPT-4o Mini and Gemini 3.1 Flash Lite (Stable).
**Messaging**\ When enabled, the AI can use interactive buttons, list menus, and product carousels, allowing customers to select options with a tap instead of typing.
**Read & Process Files & Images** \ Allows the AI Agent to process supported files and images shared by customers.
Enable this option to allow the AI Agent to read and process supported files and images shared by customers.The maximum file size is limited to 10 MB. **Summarize the Conversation**\ Generates a summary of the conversation as an Attribute. You can save it to the CRM, pass it to external software, or use it as context for another AI Agent.
**Conversation Limit** Ends the conversation when the configured message limit is reached or a specified keyword is detected.
Use **Conversation Limit** to control when the AI Agent should automatically end a conversation. **End the conversation after this many customer messages:** Set the maximum number of messages the customer can send while the AI Agent is active. Once the limit is reached, the conversation will be ended automatically. **End if no reply for:** Set a time limit for customer inactivity. If the customer does not reply within the configured time, the AI Agent will automatically end the conversation. **End the conversation when the customer sends any of these keywords:** Add specific keywords that indicate the customer wants to end the conversation. eg:- stop,cancel. ### Test AI Agents Open **Test Agent** from **Configure Agent**, or select **Via TestBot** on the flow canvas. Have a real conversation with your configuration.
The test panel shows the active model, cost per turn, and token breakdown. | Metric | Meaning | | ---------------- | ------------------------------------------------------------------------------------------------------------ | | Input tokens | Everything sent to the model, including Instructions, Knowledge Base, data schema, and conversation history. | | Output tokens | The length of the agent response. Detailed responses cost more. | | Reasoning tokens | Extra billed tokens the model uses before answering. | | Time | The model generation time for that turn. | {% hint style="info" %} WhatsApp delivery typically adds two to four seconds beyond the Test Agent time. This is normal delivery overhead and does not affect AI cost. {% endhint %} **Estimate AI cost** Each AI Agent reply is billed by the selected provider based on processed and generated content.
Use the Test Widget to review cost estimates for your configuration. ### Save and publish 1. Select **Save** in the **Configure Agent** window. 2. Save the flow in Flow Builder. 3. Select **Validate & Publish**. 4. Select **Publish Now** to make the flow live.
--- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation by asking a question. Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters: ``` GET https://help.pickyassist.com/setting-up-guide/ai-agent/getting-started.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.