Calute

Calute is a framework for building AI agents that can orchestrate function calls and switch between agents. It is designed for researchers experimenting with agent orchestration and function calling.

Key Features

• Agent Orchestration: Calute helps manage multiple agents and their interactions, allowing you to create complex AI systems.
• Function Calling: Calute simplifies the process of calling functions from AI agents, enabling them to interact with external tools and data sources.
• Agent Switching: Calute provides agent switching capabilities, allowing you to create more dynamic and intelligent systems that can adapt to different situations.
• Workflow Engine: Calute includes a workflow engine for defining and executing complex workflows involving LLMs and function execution.
• LLM Client Abstraction: Calute supports OpenAI and Gemini models through a unified client interface, making it easy to switch between different LLMs.

Installation

To install Calute, use the following command:

poetry install

Usage

Here are some examples of how to use Calute:

Registering an Agent

from calute import Agent, AgentFunction, Calute

def my_function(input_data: str, context_variables: dict = None):
    """My function description (LLM will read the docs!)"""
    return f"Function executed with input: {input_data}"

agent = Agent(
    id="my_agent",
    name="My Agent",
    functions=[AgentFunction(func=my_function)],
    instructions="Follow these instructions."
)

# Assuming you have an initialized OpenAI or Gemini client
# client = OpenAI(...) or client = Gemini(...)
# calute = Calute(client)
# calute.register_agent(agent)

Defining a Workflow

from calute import Workflow, WorkflowStep, WorkflowEngine, WorkflowStepType

# Assuming you have a Calute instance
# calute = Calute(...)
# workflow_engine = WorkflowEngine(calute)

# workflow = workflow_engine.create_function_workflow(
#     id="my_workflow",
#     name="My Workflow",
#     functions=[my_function]
# )

Running a Workflow

# Assuming you have a workflow defined
# result = await workflow_engine.execute_workflow(
#     workflow_id="my_workflow",
#     context_variables={"input_data": "Hello, world!"}
# )
# print(result)

Modules

• calute.py: Core Calute class with orchestration capabilities.
• chain_module.py: Function chaining and execution.
• client.py: LLM client abstraction for OpenAI and Gemini.
• executors.py: Function and agent execution logic.
• workflow.py: Workflow engine for defining and executing complex workflows.
• types/: Data types and structures used in Calute.
• utils.py: Utility functions.
• basics.py: Basic registries.

Types

• Agent: Represents an AI agent with capabilities and functions.
• PromptTemplate: Configurable template for structuring agent prompts.
• FunctionCall: Represents a function call to be executed.
• Workflow: Defines a workflow with steps and transitions.

Example Mermaid Diagram

graph LR
    A[User Prompt] --> B{Agent Orchestrator}
    B --> C{Agent 1}
    C --> D[Function Call 1]
    D --> E{Function Executor}
    E --> F[Result 1]
    F --> B
    B --> G{Agent 2}
    G --> H[Function Call 2]
    H --> E
    E --> I[Result 2]
    I --> J[Final Response]

Loading

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Contributing

Contributions are welcome! Please see the contributing guidelines for more information.

Contact

If you have any questions or issues, please contact erfanzar.