Customer Service AI Agent

LangChain-based customer support agent with:

• Retrieval-augmented responses from a local knowledge base (knowledge/*.txt, optional PDFs)
• Optional backend API tooling for order/customer workflows
• Two local interfaces: CLI and Streamlit web chat

Repository: https://github.com/Stennis1/Customer-Service-AI-Agent Live URL: https://customer-service-ai-agent-a97r.onrender.com/

Features

• Tool-enabled agent (search_knowledge_base, check_order_status, escalate_to_human)
• Conversation memory (last 10 exchanges)
• Chroma vector store for semantic document retrieval
• Optional REST API integration for live order status/tickets/customers
• CLI mode for terminal usage
• Web mode for browser chat via Streamlit

Tech Stack

• Python 3.10+
• LangChain
• OpenAI API
• ChromaDB
• Streamlit
• Requests

Project Structure

.
├── agent.py              # Core agent class and tools wiring
├── app.py                # CLI entrypoint
├── web_app.py            # Streamlit web UI entrypoint
├── knowledge_base.py     # Document loading + vector retrieval
├── api_tools.py          # Optional external API client tools
├── knowledge/            # Support knowledge corpus (.txt/.pdf)
├── requirements.txt      # Top-level dependencies
├── .env.example          # Environment variable template
└── test_setup.py         # Basic OpenAI connection check

Architecture

1. User sends a question in CLI or Web UI.
2. CustomerServiceAgent in agent.py receives the message.
3. Agent selects tools when needed:
   • search_knowledge_base -> semantic search via Chroma in knowledge_base.py
   • check_order_status -> external API (api_tools.py) if configured, else mock fallback
   • escalate_to_human -> escalation response
4. Agent returns final answer to the interface.

Prerequisites

• Python 3.10 or newer
• OpenAI API key
• pip and virtual environment support

Setup

1. Clone and enter project:

git clone https://github.com/Stennis1/Customer-Service-AI-Agent.git
cd Customer-Service-AI-Agent

2. Create and activate a virtual environment:

python3 -m venv .venv
source .venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Configure environment:

cp .env.example .env

5. Edit .env with at least:

OPENAI_API_KEY=your_openai_key

Optional API integration:

CUSTOMER_API_BASE_URL=https://api.example.com
CUSTOMER_API_KEY=your_customer_api_key

Running Locally

CLI

python app.py

Web Interface (Streamlit)

streamlit run web_app.py

Streamlit will print a local URL (typically http://localhost:8501).

Deployment (Render)

This repository includes a Render Blueprint file:

• render.yaml in the project root

Live deployment:

• https://customer-service-ai-agent-a97r.onrender.com/

Deploy steps:

1. Push latest code to GitHub.
2. In Render, choose New -> Blueprint.
3. Select this repository (Render reads render.yaml automatically).
4. Set required environment variables in Render:
   • OPENAI_API_KEY (required)
   • OPENAI_MODEL (optional)
   • CUSTOMER_API_BASE_URL and CUSTOMER_API_KEY (optional, for live backend API calls)

Knowledge Base

• Add/update support content under knowledge/.
• Supported loaders:
  • .txt via TextLoader
  • .pdf via PyPDFLoader
• Chunks are embedded and stored in ./chromadb.

Verification

Basic OpenAI connection test:

python test_setup.py

Syntax check:

python -m py_compile agent.py app.py web_app.py api_tools.py knowledge_base.py

API Tooling Behavior

If CUSTOMER_API_BASE_URL and CUSTOMER_API_KEY are set:

• check_order_status uses live HTTP calls.

If they are not set:

• Agent falls back to mock order-status behavior for ORD* IDs.

Environment Variables

Required:

• OPENAI_API_KEY: OpenAI API key used by model and embeddings

Optional:

• OPENAI_MODEL: Chat model identifier (default: openai:gpt-4o-mini)
• CUSTOMER_API_BASE_URL: Base URL for customer/order/support backend
• CUSTOMER_API_KEY: Bearer token for backend API
• LANGCHAIN_API_KEY: For LangSmith tracing (if you enable it)
• VECTOR_DB_URL: Reserved placeholder, not currently used in runtime

Common Issues

• OPENAI_API_KEY missing:

  • Symptom: agent initialization/search errors.
  • Fix: set key in .env.

• Empty knowledge results:

  • Symptom: poor/no retrieval output.
  • Fix: ensure docs exist under knowledge/ and restart app.

• API requests fail:

  • Symptom: order/customer/ticket tool errors.
  • Fix: verify CUSTOMER_API_BASE_URL, CUSTOMER_API_KEY, backend reachability.

Development Notes

• Keep app.py and web_app.py as thin entrypoints; put logic in agent.py.
• knowledge_base.py provides a backward-compatible alias KnowledegeBase for legacy typoed imports.
• Prefer adding tests before expanding tools/agent prompt complexity.

Next Improvements

• Add pytest tests for:
  • Tool behavior
  • Fallback paths when env vars are missing
  • Knowledge retrieval quality
• Add structured logging and request tracing
• Add Dockerfile and compose profile for reproducible local runtime
• Add CI workflow (lint + test + type checks)

License

No license file is currently included. Add LICENSE before public reuse/distribution.