Multi-Provider AI Agent Chatbot with OpenAI SDK

This project is an AI chatbot powered by the OpenAI Agents SDK, capable of using different LLM providers, including OpenAI and AWS Bedrock. It features multi-agent orchestration, guardrails, and a Streamlit-based interface for interaction. The project supports secure configuration via .env and seamless provider switching using LiteLLM.

🔧 Powered by OpenAI Agents SDK

This project leverages the OpenAI Agents SDK, a lightweight and powerful framework for building multi-agent workflows. The SDK simplifies the development of agentic applications by providing:

• Agents: LLMs equipped with instructions and tools.
• Handoffs: Mechanisms that allow agents to delegate tasks to other agents.
• Guardrails: Input validations and checks that run in parallel to agents.
• Function Tools: Python functions turned into tools with automatic schema generation and validation.
• Tracing: Built-in tracing to visualize and debug agentic flows.

For more details, refer to the official documentation.

🚀 Getting Started

---

Features

• Chatbot Interface: A user-friendly chatbot built with Streamlit.
• Agent Management: Backend logic for managing multiple agents, including a triage agent and specialized agents.
• Guardrails: Input validation and routing using guardrails.
• Asynchronous Execution: Supports asynchronous operations for efficient processing.
• Multi-Provider Support: Automatically switches between OpenAI and AWS Bedrock (Anthropic Claude 3) using LiteLLM, based on available environment credentials.
• Testing: Includes unit tests for backend logic using pytest and pytest-asyncio.
• Custom Tools with Function Calling: Create Python functions (like fetching ChatGPT IP whitelist) and expose them to LLMs using @function_tool.
• File Retrieval with Vector Store (Optional): Retrieve relevant information from your uploaded documents using OpenAI’s FileSearchTool.
  If no VECTOR_STORE_ID is provided, the agent will still function normally using only the base model.

Model Provider Flexibility with LiteLLM

The sa_genai_agent is designed to dynamically select between different LLM providers:

• If MODEL_PROVIDER is set to "bedrock" and AWS_ACCESS_KEY_ID and related AWS credentials are found in your environment or .env, it uses Anthropic Claude 3 Sonnet via AWS Bedrock.
• If not, it defaults to OpenAI’s GPT-4o model — no configuration changes needed.

This is enabled using the LiteLLM integration, making the agent flexible and cost-efficient depending on your cloud environment.

Example .env

OPENAI_API_KEY=your_openai_api_key_here
VECTOR_STORE_ID=your_vector_store_id_here (optional)
AWS_ACCESS_KEY_ID=your_aws_access_key_id_here (optional)
AWS_SECRET_ACCESS_KEY=your_aws_secret_access_key_here (optional)
AWS_REGION_NAME=eu-west-1 (optional)
MODEL_PROVIDER=openai  # "openai" or "bedrock"

---

Installation

1. Clone the repository:

   git clone https://github.com/your-repo/openAiAgentText.git
   cd openAiAgentText

2. Install dependencies:

   pip3 install -r requirements.txt

3. Create a .env file in the root directory and add your credentials. The app will use OpenAI by default, but if AWS credentials are present, it will use AWS Bedrock with the Anthropic Claude 3 model:

   OPENAI_API_KEY=your_openai_api_key_here
   VECTOR_STORE_ID=your_vector_store_id_here (optional)
   AWS_ACCESS_KEY_ID=your_aws_access_key_id_here (optional)
   AWS_SECRET_ACCESS_KEY=your_aws_secret_access_key_here (optional)
   AWS_REGION_NAME=eu-west-1 (optional)
   MODEL_PROVIDER=openai  # "openai" or "bedrock"

   • To obtain a VECTOR_STORE_ID (optional – the app will still work even if you don't provide a vector store):

     • Upload files via the OpenAI Platform UI:

       1. Navigate to OpenAI's platform.
       2. Go to the "Files" section.
       3. Create a new vector store and attach your uploaded file.
       4. Retrieve the VECTOR_STORE_ID from the vector store details.

     • Or use the OpenAI CLI:

       openai file create -p assistants -f my_document.pdf
       openai vector-store create -f <file_id>

---

Usage

Run the Chatbot

To execute the Chatbot logic:

python3 -m streamlit run frontend/chatbot_app.py

To execute the backend logic (optional):

python3 backend/main.py

Run Tests

To execute the test suite:

python3 -m pytest tests/test_main.py

To save test results to a file:

python3 -m pytest tests/test_main.py > test_results.txt

---

🤖 Agent Routing Examples

These examples show how user inputs are routed to the correct agent or rejected, based on the triage logic and guardrails.

💬 Example 1

User Input: How can I integrate Amazon Bedrock with LangChain?

• 🔍 Summary: A question about using generative AI tools and orchestration frameworks.
• 🤖 Routed Agent: SA Generative AI Specialist (sa_genai_agent)

💬 Example 2

User Input: How can I automate deployment pipelines using GitHub Actions and AWS CodePipeline?

• 🔍 Summary: A DevOps/operations-related question about CI/CD automation.
• 🤖 Routed Agent: SA Operations Specialist (sa_operations_agent)

💬 Example 3

User Input: Regarding my Generative AI Chatbot architecture, I need to whitelist an IP address for ChatGPT. Could you provide one?

• 🔍 Summary: Generative AI architecture with a specific technical requirement (ChatGPT IPs). will call a Custom Tool (chatgpt_actions_tool.py)
• 🤖 Routed Agent: SA Generative AI Specialist (sa_genai_agent)

💬 Example 4

User Input: What is 1 + 1?

• 🔍 Summary: A general question not related to architecture.
• ❌ Routed Agent: None – blocked by guardrail (is_architecture: false)

---

🔌 Custom Tool Example

This project includes a custom tool (get_chatgpt_actions) that uses a public OpenAI API to fetch the latest ChatGPT IP whitelist.

The tool is implemented as a Python function and exposed to the LLM via the @function_tool decorator.

---

Project Structure

openAiAgentsChatbot/
├── backend/                    # Backend logic
│   ├── agents/                 # Agent definitions
│   │   ├── sa_genai_agent.py   # Generative AI specialist agent
│   │   ├── sa_operations_agent.py # Operations specialist agent
│   ├── tools/
│   │   ├── chatgpt_actions_tool.py  # Tool to fetch ChatGPT IP whitelist
│   ├── prompts/                # agent instructions
│   │   ├── guardrails.yaml
│   │   ├── sa_operations_instructions.yaml
│   │   └── sa_genai_agent_instructions.yaml
│   │   └── triage_agent_instructions.yaml
│   ├── [main.py](http://_vscodecontentref_/3) # Backend entry point
├── frontend/                   # Frontend logic
│   ├── [chatbot_app.py](http://_vscodecontentref_/4) # Streamlit chatbot application
├── tests/                      # Unit and integration tests
│   ├── test_main.py            # Tests for backend logic
├── .env                        # Environment variables (ignored by Git)
├── requirements.txt            # Python dependencies
├── [README.md](http://_vscodecontentref_/5) # Project documentation

---

Dependencies

Install the following Python packages:

• openai-agents
• python-dotenv
• pyyaml
• openai
• streamlit
• pytest
• pytest-asyncio

You can install all dependencies using:

pip3 install -r requirements.txt

or one by one:

pip3 install openai-agents  
pip3 install python-dotenv
pip3 install pyyaml    
pip3 install openai streamlit
pip3 install pytest
pip3 install pytest-asyncio
pip3 install "openai-agents[litellm]"
pip3 install litellm
pip3 install boto3

Author

Developed by imKikev.

---

Key Improvements

1. Detailed Usage Instructions: Includes commands for running the backend, chatbot, and tests.
2. Project Structure: Provides an overview of the folder structure for better understanding.
3. Environment Variables: Explains how to configure the .env file.
4. Testing: Includes commands for running tests and generating reports.
5. Troubleshooting: Addresses common issues like missing environment variables or dependencies.

Let me know if you'd like to customize this further!

---

License

This project is licensed under the MIT License. See the LICENSE file for details.