shaheenai 0.1.1

A flexible, multi-LLM, agent-oriented Python library with self-reflection capabilities

shaheenai

ShaheenAI is a flexible, multi-LLM, agent-oriented Python library that supports multiple language model providers like OpenAI, Anthropic, Ollama, and Cohere via a plugin/extras architecture. The library offers self-reflection, tool invocation, task chaining, and optional UI integrations using Streamlit and Chainlit.

Features

• Modular Agent Class: Supports multiple LLMs with self-reflection, tool invocation, and task chaining.
• MCP Server Interface: Optionally integrates tools and external APIs.
• Configurable via YAML or Code: Supports playbooks and programmatic configuration.
• Wide LLM Provider Support: OpenAI, Anthropic, Ollama, Cohere, etc.
• Streamlit and Chainlit Support: Build interactive and conversational UIs for agents.

Getting Started

Prerequisites

• Python 3.10 or higher

Installation

To install ShaheenAI, use pip:

pip install shaheenai

Usage

Here's a basic example of how to use ShaheenAI to create an agent that utilizes a weather tool:

from shaheenai.agent import Agent
from shaheenai.mcp import MCP

# Define and run the MCP server
mcp = MCP()
@mcp.tool()
async def get_weather(location: str) -> str:
    return "Sunny with 25°C"

if __name__ == "__main__":
    mcp.run()

# Create an agent
agent = Agent(instructions="You can use tools when needed", llm="openai", tools=["get_weather"])
response = agent.start("What's the weather in Lahore today?")
print(response)

CLI

ShaheenAI provides a command-line interface for running agents defined in YAML playbooks or via auto-mode.

Example:

shaheenai run agents.yaml

Directory Structure

shaheenai/
 ├── shaheenai/
 │    ├── __init__.py
 │    ├── agent.py
 │    ├── mcp.py
 │    ├── llm_providers/
 │    │     ├── openai.py
 │    │     ├── cohere.py
 │    │     └── ...
 │    ├── tools/
 │    └── config.py
 ├── setup.py or pyproject.toml
 ├── README.md
 ├── LICENSE  # e.g. MIT
 └── examples/
      ├── app.py
      └── agents.yaml

Contributing

Contributions are welcome! Please read the contribution guidelines first.

License

This project is licensed under the MIT License.

Acknowledgments

Inspired by PraisonAI for its modularity and multi-LLM support.