Skip to content

README.md

Latest commit

 

History

History
121 lines (82 loc) · 3.51 KB

README.md

File metadata and controls

121 lines (82 loc) · 3.51 KB
Raw

🤖 Support Agent System Utilizing the OpenAI Agents SDK

A multi-agent console support system built using the OpenAI Agents SDK and Gemini API (gemini-2.5-flash). This system routes user queries to specialized agents (billing, technical, general) with dynamic tools, handoffs, output guardrails, and live streaming interactions.

🧠 Project Objective

Build a responsive, intelligent support system that leverages:

  • ✅ Multi-agent collaboration to resolve diverse support queries
  • ✅ Intelligent triage and routing via the Triage Agent
  • ✅ Dynamic tool execution based on user context (e.g., premium access)
  • ✅ Guardrails to ensure professionalism in replies
  • ✅ Live interaction with streaming outputs for transparency

🚀 Features

Feature Description
🔁 Triage Agent Classifies user intent and delegates to specialist agents
🧾 Billing Agent Handles refunds, invoices, subscriptions
🛠️ Technical Agent Resolves restarts, errors, and connectivity issues
💬 General Agent Answers general or unclassified questions
🛑 Guardrail Agent Blocks apologetic tone (e.g., no “sorry”)
🧰 Smart Tools Tools like refund() or restart_service() only activate when conditions are met
🧵 Streaming Output Shows responses, handoffs, and tool actions in real time

🛠️ Technologies Used

Component Description
🐍 Python Base language for async and logic
🧠 OpenAI Agents SDK Multi-agent framework for building AI agents
🔮 Gemini API Model: gemini-2.5-flash for completions
🧱 Pydantic Input/output schema validation
🔐 dotenv Environment variable management
🔁 asyncio Asynchronous event loop support
⚡ uv Fast Python package manager & virtualenv tool

🧪 Setup Instructions with uv

🧰 Note: This project uses uv — a modern, blazing-fast Python environment manager.

🧷 Step 1: Initialize your environment

uv init

📦 Step 2: Add required dependencies

uv add openai-agents

🔐 Step 3: Activate the environment

.venv\Scripts\activate  # Windows
# OR
source .venv/bin/activate  # macOS/Linux

🔑 Step 4: Add your Gemini API key

Create a .env file in the root of the project and add:

GEMINI_API_KEY=your-gemini-api-key

🏃 Step 5: Run the console app

uv run main.py

💡 Example Interaction

🔧 Support System Initialized — type 'exit' to quit.
👤 Enter your name: Ali
⭐ Premium user? (y/N): y

💬 How can I help you? I need a refund.

[ 🧠 Triage in progress… ]
→ Handoff: Billing Agent
→ Tool Call: refund
✅ Refund for Ali initiated; expect 3–5 business days.

📌 Key Notes

  • 🛑 Apology guardrail ensures professional tone by blocking words like “sorry”.
  • 🧰 Contextual tools: Tools only activate if user conditions are met (e.g., refund is for premium users).
  • 🧵 Streamed feedback: Real-time console output of all steps—tool calls, handoffs, and agent messages.