> ## Documentation Index > Fetch the complete documentation index at: https://docs.subverseai.com/llms.txt > Use this file to discover all available pages before exploring further. # AI Query Agent Node > Automate intelligent decision-making with AI-powered processing ## Overview The AI Query Agent Node enables automated decision-making in your workflows. It uses AI to analyze data, make intelligent decisions, and perform multiple actions based on dynamic conditions, mimicking human-like intelligence. ## Use Cases * **Post-Call Processing**: Analyze call transcripts and determine next actions * **Smart Routing**: Decide which team or agent should handle a case * **Ticket Generation**: Create support tickets with AI-generated summaries * **Email Composition**: Generate personalized emails based on context * **Lead Qualification**: Score and categorize leads automatically * **Callback Scheduling**: Intelligently determine if and when to call back * **Data Extraction**: Pull specific information from unstructured text * **Sentiment Analysis**: Determine customer satisfaction and urgency * **Action Orchestration**: Coordinate multiple follow-up actions ## How It Works 1. **Input**: Provide a query/instruction and context (e.g., call transcript) 2. **Processing**: AI analyzes the context using selected tools and knowledge base 3. **Decision**: AI determines appropriate actions based on your query 4. **Execution**: AI calls selected tools to perform actions 5. **Output**: Results and tool outputs are passed to next workflow blocks ## Inputs ### **Query** (Required) User input or query that instructs the AI on what to do. **Purpose:** * Define the task for the AI * Specify conditions and logic * Describe desired outcomes **Example Queries:** **Post-Call Actions:** ```md theme={null} 1. Generate a ticket with comments as per discussion and assign to relevant department 2. Send an email to the customer with ticket details, CC to customercare@company.com 3. If the customer asked to call later / callback, schedule a callback accordingly 4. If the customer was willing to speak but the call got hung up abruptly, schedule a callback in 5 minutes ``` **Lead Qualification:** ```md theme={null} Analyze the conversation and: 1. Rate the lead quality from 1-10 2. If score is above 7, send lead details to sales team via webhook 3. If score is below 4, send a follow-up email with more information 4. Extract and return: customer name, company, budget, timeline ``` **Customer Support:** ```md theme={null} Based on the call transcript: 1. Determine if the issue was resolved 2. If not resolved, create a support ticket with summary 3. If customer mentioned specific product, query knowledge base for details 4. Send follow-up email with resolution or ticket number ``` **Smart Routing:** ```md theme={null} Analyze the customer inquiry and: 1. Categorize as: Sales, Support, Billing, or Technical 2. Determine urgency level: High, Medium, Low 3. If High urgency, send immediate alert via webhook 4. Route to appropriate department ``` ### **Tools** Select the tools/functions the AI can use to perform actions. **Available Tools:** * **send\_email**: Send emails to customers * **send\_whatsapp**: Send WhatsApp messages * **schedule\_callback**: Schedule follow-up calls * **create\_ticket**: Generate support tickets * **query\_knowledge\_base**: Search knowledge base * **webhook**: Send data to external systems * **Custom tools**: Your configured LLM tools **Selection Tips:** * Only enable tools needed for the task * More tools = more processing time * Test with minimal tools first * Add tools as needed ### **System\_Context** Provide additional context to help the AI make informed decisions. **What to Include:** * Call transcripts * Customer details * Previous interactions * Order information * Account status * Any relevant data **Example Context:** **Call Transcript:** ```md theme={null} Call Transcript: Assistant: Hello! How can I help you today? Customer: Hi, I placed an order yesterday but haven't received confirmation. Assistant: I can help with that. Can you provide your order number? Customer: Yes, it's ORD-12345. Assistant: Let me check... I see your order. It's confirmed and will ship tomorrow. Customer: Great, thank you! Customer Details: Name: John Doe Email: john@example.com Phone: +91xxxxxxxxxx Order ID: ORD-12345 ``` **CRM Data:** ```md theme={null} Customer: Jane Smith Account Type: Premium Last Purchase: 2025-11-15 Total Spend: ₹50,000 Support Tickets: 2 (both resolved) Sentiment: Positive ``` **System Prompt:** ```md theme={null} You are a backend agent for SubVerse AI to perform multiple tasks based on customer's discussion with virtual agent. Guidelines: - Be concise and professional - Extract key information accurately - Make decisions based on clear criteria - Use appropriate tools for each action - Provide clear summaries ``` ### **Knowledge\_Base** Select a knowledge base to pull relevant information. **Use Cases:** * Product information lookup * Policy clarification * Technical documentation * FAQ answers * Company procedures **Configuration:** * Select from your created knowledge bases * AI will automatically search when needed * Results are included in decision-making ## Outputs ### **LLM\_Response** The AI's response based on your query and context. **Contains:** * Analysis summary * Decisions made * Actions taken * Extracted information * Recommendations **Example Output:** ```md theme={null} Analysis: - Customer inquiry about order status - Order ORD-12345 is confirmed and shipping tomorrow - Customer is satisfied with the response - No further action needed Extracted Information: - Customer Name: John Doe - Order ID: ORD-12345 - Issue: Order confirmation inquiry - Resolution: Confirmed order status - Sentiment: Positive Actions Taken: - Sent confirmation email to john@example.com - Updated CRM with interaction details ``` ### **Tool Outputs** Results from tools that were called by the AI. **Available Outputs (based on tools used):** * **email\_sent**: Boolean (true/false) * **whatsapp\_sent**: Boolean (true/false) * **callback\_scheduled**: Boolean (true/false) * **ticket\_id**: String (ticket number) * **webhook\_response**: Object (API response) * **knowledge\_base\_results**: Array (search results) **Usage:** * Connect to subsequent workflow blocks * Use for conditional logic * Log for monitoring * Send to external systems ## Configuration Examples ### Example 1: Post-Call Email Follow-up **Query:** ```md theme={null} Analyze the call transcript and: 1. Summarize the conversation in 2-3 sentences 2. If customer requested information, query knowledge base 3. Send a follow-up email with summary and any requested information ``` **Tools:** `send_email`, `query_knowledge_base` **System\_Context:** `[Call Transcript]` ### Example 2: Support Ticket Creation **Query:** ```md theme={null} Based on the call: 1. Determine if issue was resolved 2. If not resolved, create a support ticket with: - Customer details - Issue description - Priority level - Assigned department 3. Send email to customer with ticket number ``` **Tools:** `create_ticket`, `send_email` **System\_Context:** `[Call Transcript + Customer Details]` ### Example 3: Lead Qualification **Query:** ```md theme={null} Analyze the conversation and score the lead: 1. Rate interest level (1-10) 2. Extract: budget, timeline, decision maker 3. If score > 7, send to sales via webhook 4. If score < 4, schedule follow-up call in 1 week 5. Send summary email to lead ``` **Tools:** `webhook`, `schedule_callback`, `send_email` **System\_Context:** `[Call Transcript + Lead Details]` ### Example 4: Smart Callback Scheduling **Query:** ```md theme={null} Determine if callback is needed: 1. If customer explicitly requested callback, schedule at requested time 2. If call dropped abruptly, schedule callback in 5 minutes 3. If issue unresolved, schedule callback in 24 hours 4. Otherwise, no callback needed ``` **Tools:** `schedule_callback` **System\_Context:** `[Call Transcript + Call Metadata]` ## Best Practices ### Query Design * **Be Specific**: Clearly define what you want the AI to do * **Use Numbers**: List actions in numbered steps * **Set Conditions**: Use if/then logic for decisions * **Define Criteria**: Specify thresholds and rules * **Keep Focused**: One clear objective per query ### Context Provision * **Include Relevant Data**: Only what's needed for the decision * **Structure Clearly**: Use labels and formatting * **Update Regularly**: Ensure context is current * **Validate Data**: Check for completeness ### Tool Selection * **Minimal Set**: Only enable necessary tools * **Test Individually**: Verify each tool works * **Monitor Usage**: Track which tools are called * **Optimize**: Remove unused tools ### Knowledge Base * **Keep Updated**: Regularly refresh content * **Organize Well**: Structure for easy retrieval * **Test Queries**: Verify search accuracy * **Monitor Relevance**: Check result quality ## Advanced Use Cases ### Multi-Step Workflows **Scenario:** Complex post-call processing ```md theme={null} Query: 1. Analyze call sentiment (positive/negative/neutral) 2. Extract key information: issue, resolution, follow-up needed 3. If negative sentiment: - Create high-priority ticket - Alert supervisor via webhook - Schedule manager callback within 2 hours 4. If positive sentiment: - Send thank you email - Request feedback via WhatsApp 5. Update CRM with all details ``` ### Conditional Routing **Scenario:** Department-based routing ```md theme={null} Query: Categorize the inquiry and route appropriately: 1. If billing-related: send to finance via webhook 2. If technical issue: create tech support ticket 3. If sales inquiry: send to sales team and schedule callback 4. If general question: query knowledge base and send email response ``` ### Data Enrichment **Scenario:** Customer profile enhancement ```md theme={null} Query: Enrich customer profile based on conversation: 1. Extract: interests, pain points, budget range, decision timeline 2. Update CRM via webhook with extracted data 3. Calculate lead score based on extracted information 4. Assign to appropriate sales rep based on score and territory ``` ## Monitoring and Optimization ### Track Performance **Key Metrics:** * **Execution Time**: How long AI takes to process * **Tool Usage**: Which tools are called most * **Success Rate**: Percentage of successful actions * **Error Rate**: Failed tool calls or decisions ### Analyze Outputs **Review:** * LLM response quality * Decision accuracy * Tool call appropriateness * Context utilization ### Iterate and Improve **Optimization:** * Refine queries based on results * Adjust tool selection * Improve context structure * Update knowledge base ## Troubleshooting ### AI Not Calling Tools **Possible Causes:** * Tools not selected in configuration * Query doesn't clearly instruct tool usage * Insufficient context for decision **Solutions:** * Verify tools are enabled * Make query more explicit about tool usage * Provide more detailed context ### Incorrect Decisions **Possible Causes:** * Ambiguous query * Insufficient context * Outdated knowledge base **Solutions:** * Clarify query with specific criteria * Include all relevant context * Update knowledge base content ### Slow Processing **Possible Causes:** * Too many tools enabled * Large context size * Complex query **Solutions:** * Reduce number of tools * Optimize context length * Simplify query ## Next Steps Configure email sending Set up callback scheduling Integrate with external systems Create custom tools Start with simple queries and gradually add complexity. Test each step to ensure the AI makes correct decisions. The AI Query Agent uses the LLM model configured in your account. Processing time depends on query complexity and context size.