Workflow Description: Scalable and Automated Business Email & Chat Handling

This workflow is designed to automate and scale the processing, classification, and routing of business emails and chat messages, with a particular focus on customer support scenarios. It uses a combination of AI-powered classification, conditional logic, API integrations, and context-aware agent responses, making it highly useful for organizations looking to streamline customer support, detect spam, and integrate AI agents with external systems like LiveAgent.

High-Level Overview

The flow consists of two main branches:

1. Email/Message Spam Detection and Routing: Automatically classifies incoming messages as spam or not and routes them accordingly to ensure efficient handling and reduce manual filtering.
2. Contextual AI Agent for Customer Support: Leverages external document retrieval and a sophisticated tool-calling agent to provide accurate, context-rich, and policy-compliant customer support responses.

Detailed Flow Structure

1. Spam Detection and Routing

• Input: Messages are received through a chat input node.
• Prompt Construction: A prompt template is filled with the incoming message and sent to an API request node (configured for LiveAgent or similar systems).
• Text Extraction: The API response is parsed to extract the relevant content using a parsing node.
• Spam Classification LLM: The extracted message is sent to an OpenAI GPT-4.1 model with a system prompt that instructs it to output “yes” if the message is spam (e.g., promotions, unsolicited offers) or “no” if it is business-relevant.
• Conditional Routing:
  • If the message is classified as spam (“yes”), a prompt template generates a standard spam notification, which is output to chat.
  • If the message is not spam (“no”), it proceeds to a context-aware AI support agent for further handling.

Spam Routing Logic Table:

2. Contextual AI Customer Support Agent

• Document Retrieval: When a message is not spam, the flow retrieves relevant documents from a knowledge base, filtered by categories and schedules, to ensure up-to-date and contextually appropriate information is available.
• Chat History Integration: Recent chat history is retrieved to give the agent context for its responses, ensuring continuity and relevance.
• LLM Agent Orchestration: An Anthropic Claude Sonnet 4 model powers a Tool Calling Agent, which is given a detailed system message outlining strict response policies (e.g., always respond in English, use a polite business tone, reference URLs exactly, avoid speculation, and follow a specific output format).
• Tool Integration: The agent can call the document retriever tool and other APIs as needed to supplement its responses, ensuring fact-based answers.
• Output: The agent’s response is output to chat and also passed along for further API integration if required (e.g., forwarding to LiveAgent).

Agent Response Logic Table:

System Message Guidelines for Agent

• Always respond in English regardless of the input language.
• Use a formal but friendly business tone; no markdown, no emojis, and no speculation.
• Responses must cite URLs directly below the answer and follow a clear, bullet-friendly structure.
• If no relevant info is found, politely ask the customer for more details.

3. API Integration and Data Handling

• Dynamic Data Creation: Several nodes generate the required data dictionaries for API requests, including API keys, message bodies, and query parameters.
• Multiple API Interactions: The workflow accommodates multiple API requests (e.g., to LiveAgent, other backend systems) with flexible methods (GET, POST, etc.).
• Data Parsing: After each API response, parsing nodes extract relevant text for downstream processing or output.

4. Workflow Automation and Scaling Benefits

• Automated Spam Filtering: Eliminates manual review of unsolicited messages, saving time and focusing agent attention on real support requests.
• Consistent, Policy-Compliant Support: The agent always follows company guidelines, ensuring reliable, brand-safe communication.
• Knowledge-Driven Responses: By integrating document retrieval, the agent bases its answers on the latest and most relevant information.
• Flexible API Integration: Easily extendable to new tools, ticketing systems, or knowledge sources as business needs evolve.
• Scalable: The modular design enables parallel processing of multiple conversations and easy adaptation to higher message volumes.

Example Use Cases

• Customer Support Teams: Automatically triage messages, handle spam, and provide accurate, context-rich responses.
• Sales/Lead Qualification: Filter out non-relevant inquiries before passing genuine leads to sales reps.
• Omnichannel Integration: Connects with platforms like LiveAgent or custom APIs to unify support across channels.

Setup Notes

• Customizations Required: Replace placeholder API keys and URLs (e.g., “YOURLINK”) with actual deployment values.
• Knowledge Base Management: Keep document retriever sources updated for optimal agent performance.

This workflow provides a robust, scalable architecture for automating customer communication, ensuring both efficiency and high-quality, policy-compliant responses. Its modularity allows for easy expansion as your business grows or as new automation needs arise.