β οΈ Warning: Since agentic capabilities are evolving fast, expect the API to change.
A lightweight Python library for building AI agents. Turn any Python function or class into AI tools. Supports OpenAI, Anthropic, async operations, and streaming conversations.
- Add functions:
agent.add_function(my_function)
- Add classes:
agent.add_tool(my_class_instance)
- MCP support
- Multiple providers: OpenAI, Anthropic, ...
- Async/await support
- Image processing
- Conversation streaming
- Template system
Agentlys excels at building development agents. Here's how agentlys-dev uses agentlys to create an AI developer:
from agentlys import Agentlys
# pip install 'agentlys-tools[all]'
from agentlys_tools.code_editor import CodeEditor
from agentlys_tools.terminal import Terminal
# Create a developer agent
agent = Agentlys(
instruction="""You are a developer agent equipped with tools to:
1. Edit code files
2. Run terminal commands
3. Test and debug applications""",
provider="anthropic",
model="claude-sonnet-4-20250514",
name="Developer"
)
# Add development tools
code_editor = CodeEditor()
agent.add_tool(code_editor)
terminal = Terminal()
agent.add_tool(terminal)
# The agent can now autonomously develop, test, and deploy code
for message in agent.run_conversation("Create a FastAPI hello world app with tests"):
print(message.to_markdown())
Install agentlys with all providers and features:
pip install 'agentlys[all]'
Or install with specific providers:
# OpenAI only
pip install 'agentlys[openai]'
# Anthropic only
pip install 'agentlys[anthropic]'
# With MCP support (Python 3.10+)
pip install 'agentlys[mcp]'
Turn regular Python functions into tools by using add_function()
- Methods docstring, args and return type will be used to generate the tool description.
from agentlys import Agentlys
def get_weather(city: str) -> str:
return f"Sunny in {city}"
agent = Agentlys()
agent.add_function(get_weather)
agent.ask("What's the weather in Tokyo?")
Turn entire classes into tools by using add_tool()
- Methods docstring, args and return type will be used to generate the tool description.
- __llm__ method will be used to give AI the last state of the tool at each interaction.
import os
class FileManager:
def __llm__(self):
return "Files:\n" + "\n".join(os.listdir(self.directory))
def read_file(self, path: str) -> str:
"""Read a file
Args:
path: Path is relative to the directory or absolute
"""
with open(path) as f:
return f.read()
def write_file(self, path: str, content: str):
with open(path, 'w') as f:
f.write(content)
file_manager = FileManager()
agent = Agentlys()
agent.add_tool(file_manager)
# AI can now read/write files
for msg in agent.run_conversation("Read config.json and update the port to 8080"):
print(msg.content)
from agentlys import Agentlys, Message
from PIL import Image
agent = Agentlys()
image = Image.open("examples/image.jpg")
message = Message(role="user", content="Describe this image", image=image)
response = agent.ask(message)
# Load agent from markdown template
agent = Agentlys.from_template("./agent_template.md")
# Async operations
response = await agent.ask_async("Hello")
async for message in agent.run_conversation_async("Help me code"):
print(message.content)
# Set up your API keys
export OPENAI_API_KEY="your-key"
export ANTHROPIC_API_KEY="your-key"
# Choose your model (optional)
export AGENTLYS_MODEL="claude-sonnet-4-20250514" # or gpt-5-mini
π‘ Recommendation: Use Anthropic's Claude models for complex agentic behavior and tool use.
- π€ AI Assistants: Build conversational assistants with tool access
- π οΈ Development Agents: Create agents that can code, test, and deploy (like agentlys-dev)
- π Data Analysis: Agents that can query databases, generate reports, visualize data
- π Web Automation: Agents that interact with web APIs and services
- π Task Automation: Automate complex workflows with AI decision-making
- π― Custom Tools: Integrate your existing Python tools with AI
- API Reference - Complete API documentation
- Examples - More example implementations
- Provider Guide - Working with different LLM providers
- Tool Development - Creating custom tools
- Best Practices - Tips for building robust agents
If you encounter any issues or have questions, please file an issue on the GitHub project page.
This project is licensed under the terms of the MIT license.