AgentOps Lite

Lightweight agentic AI system demonstrating LLM tool-calling, enterprise workflow automation, and observable agent execution patterns.

Overview

AgentOps Lite is a prototype AI agent designed to simulate enterprise customer support workflows. The system uses an LLM-powered decision engine to dynamically select and execute tools, enabling action-oriented responses rather than traditional chatbot-style interactions.

The project demonstrates core concepts behind modern agentic systems, including:

• LLM tool calling
• Multi-step reasoning loops
• Enterprise workflow automation
• Structured agent observability
• CRM-style support operations

Architecture

            ┌──────────────────────┐
            │   User Query         │
            └─────────┬────────────┘
                      │
                      ▼
        ┌───────────────────────────┐
        │   LLM Agent               │
        │  - Reasoning              │
        │  - Tool Selection         │
        └─────────┬─────────────────┘
                  │
                  ▼
     ┌──────────────────────────────┐
     │  Tool Execution Layer        │
     │  - Order Lookup              │
     │  - Refund Processing         │
     │  - Ticket Creation           │
     └─────────┬────────────────────┘
               │
               ▼
     ┌──────────────────────────────┐
     │  Observation & Trace Layer   │
     └─────────┬────────────────────┘
               │
               ▼
     ┌──────────────────────────────┐
     │  Final Response              │
     │  + Execution Trace           │
     └──────────────────────────────┘

Features

Agentic Decision Making

The agent evaluates user requests and determines which actions should be taken using available tools.

Tool Calling

Supports dynamic execution of business functions such as:

• Order status retrieval
• Refund processing
• Support ticket creation

Observability

Every execution generates a structured trace containing:

• Tool calls
• Arguments
• Results
• Final response

Enterprise Workflow Simulation

Models common customer support workflows found in CRM and service operations environments.

Example Agent Workflow

The repository includes a complete example execution trace: examples/refund_workflow_trace.json

Agent Actions

1. Retrieve order status
2. Verify eligibility
3. Process refund
4. Generate customer response

Output

{
  "tool": "issue_refund",
  "order_id": "1001",
  "refund_status": "approved"
}

Project Structure

agentops-lite/
├── README.md
├── requirements.txt
├── .env.example
├── src/
├── data/
├── examples/
└── diagrams/

Tech Stack

• Python
• OpenAI API
• Function Calling
• JSON Tool Schemas

Future Improvements

• Multi-agent workflows
• Human-in-the-loop escalation
• Vector database memory
• Web UI dashboard
• CRM integration layer

Motivation

This project was built to explore how modern AI agents move beyond conversational interfaces and interact directly with business workflows through structured tool execution and reasoning loops.

Note :

The demo uses the Gemini free tier, so API rate limits may apply during repeated testing.