Workflow Description

This workflow is designed to automate and scale advanced customer support and knowledge retrieval tasks, making use of LLMs (Large Language Models), dynamic data creation, external API requests (such as LiveAgent), and automated document retrieval. It is especially useful for organizations seeking to streamline support processes, respond to customer queries with context-aware responses, and integrate knowledge-base lookups with external system interactions.

High-Level Overview

The workflow orchestrates these main steps:

• Receives user input (via chat)
• Builds dynamic API requests based on user input and context
• Retrieves and parses data from external sources (e.g., LiveAgent)
• Uses LLMs to extract and summarize relevant information from responses
• Augments responses using document retrieval from a knowledge base
• Leverages an LLM-powered agent to generate customer-ready answers, always grounded in retrieved context
• Presents the response back to the user

Main Components and Flow

Detailed Steps

1. User Input and Context Gathering

• The process begins with a Chat Input node, where the user’s message is received.
• The Chat History node retrieves the last N messages, enabling context-aware responses.
• A Prompt Template uses the user input and history to dynamically generate a URL for the external API (for example, to fetch a conversation transcript from LiveAgent).

2. API Request Construction

• Create Data nodes allow dynamic construction of query parameters or request bodies (including secure storage of API keys or other necessary fields).
• The generated URL and parameters are fed into an API Request node, which interacts with external systems (such as LiveAgent) to fetch the required data.

3. Data Parsing and Preprocessing

• API responses are processed using Parse Data nodes, transforming raw data into structured text or extracting only relevant fields.
• This parsed data is passed to the LLM Generator node, which is tasked with extracting specific information (e.g., the “Preview” section) using a well-defined system message.

4. Knowledge Augmentation

• Meanwhile, the Document Retriever node allows the system to search within internal knowledge bases for documents highly relevant to the user’s query, further enriching the agent with authoritative context. This tool is made available to the LLM agent.

5. LLM Agent Response Generation

• The Tool Calling Agent node is a powerful LLM-based agent which:
  • Receives user input, API responses, chat history, and access to tools (Document Retriever, Contact Human Assist, etc.)
  • Is guided by a detailed system prompt specifying:
    • Always use authoritative sources (e.g., Document Retriever, knowledge_source_tool)
    • Never invent answers or URLs
    • Ask clarifying questions if needed
    • Format responses in a friendly, professional, and concise manner
    • Use bullet points, bold text, and emojis for engaging responses
    • Always answer in Slovak (or detected user language), using an email tone
    • Escalate to human support if the query cannot be resolved
• This ensures that every customer response is accurate, context-based, policy-compliant, and highly scalable.

6. Output to User

• The final generated response (from the LLM agent) is parsed and formatted, then delivered to the user via Chat Output nodes.

Notes and Best Practices

• API Key and LiveAgent Link: The workflow includes note nodes reminding the user to insert their API key and replace YOURLINK in prompt templates with their actual LiveAgent instance URL.
• Security and Compliance: API keys and sensitive data are handled using dynamic data nodes, minimizing risk of accidental exposure.
• Extensibility: The modular design allows for easy addition of further tools, data transformations, or output channels.

Why Is This Workflow Useful for Scaling and Automation?

• End-to-End Automation: Integrates multiple data sources (live chat, APIs, knowledge base) and automates the decision-making and response process.
• LLM-Powered Reasoning: Leverages state-of-the-art LLMs for contextual understanding, information extraction, and human-like communication.
• Consistent, High-Quality Support: The agent’s system prompt enforces company policies, tone, escalation paths, and ensures no hallucinated data is provided.
• Rapid Integration with External Systems: Easily adapt to different APIs or knowledge bases by updating prompt templates and connection nodes.
• Human-in-the-Loop Escalation: Seamlessly hands off complex cases to human agents, ensuring coverage for edge cases.
• Scalability: Can handle large volumes of queries in parallel, with consistent accuracy and compliance.

Summary Table of Key Nodes

This workflow is ideal for automating customer support, integrating with external ticketing or chat systems, and ensuring LLM-driven responses are always grounded in authoritative company knowledge. It can be the backbone of a scalable, intelligent support assistant ready for enterprise use.