AI Agents

Build AI agent teams that work together to tackle complex tasks using Nuclia-hosted LLMs.

With the crewAI Python library and the NucliaNuaChat class, you can integrate crewAI agents with Nuclia’s LLMs.

Prerequisites

1. Install the required libraries:

   pip install crewai nuclia[litellm]

2. Obtain your Nuclia NUA key

How to Integrate crewAI Agents with Nuclia LLMs

Here’s a step-by-step guide to integrate a crewAI agent with a Nuclia-hosted LLM:

from crewai import Agent
from nuclia.lib.nua_chat import NucliaNuaChat

# Replace with your Nuclia Key
NUA_KEY = "<your-nua-key>"

# Initialize the Nuclia LLM handler
custom_llm = NucliaNuaChat(
    token=NUA_KEY,
)

# Define the LLM provider and model
PROVIDER = "nuclia"
MODEL = "claude-3"

# Map the custom LLM handler to the provider
litellm.custom_provider_map = [{"provider": PROVIDER, "custom_handler": custom_llm}]

# Define the LLM identifier
LLM = f"{PROVIDER}/{MODEL}"

# Configure and initialize the agent
Agent(
    role="Senior Software Engineer",
    goal="Create software as needed",
    backstory=(
        "You are a Senior Software Engineer at a leading tech think tank."
        "Your expertise in programming in python. and do your best to produce perfect code"
    ),
    llm=LLM,
)

A complete example

Markdown validator

This example demonstrates how to use crewAI and Nuclia-hosted LLMs to validate Markdown files. It is retrieved from the crewAI repository Markdown Validator Example.

Only modified files are shown here. For the complete code and instructions on how to run it, check the link above.

crew.py

import litellm
from crewai import Agent, Crew, Process, Task
from crewai.project import CrewBase, agent, crew, task
from nuclia.lib.nua_chat import NucliaNuaChat

from markdown_validator.tools.markdownTools import markdown_validation_tool

NUA_KEY = "<your-nua-key>"

custom_llm = NucliaNuaChat(
    token=NUA_KEY,
)

PROVIDER = "nuclia"
litellm.custom_provider_map = [{"provider": PROVIDER, "custom_handler": custom_llm}]

LLM = f"{PROVIDER}/claude-3"

@CrewBase
class MarkDownValidatorCrew():
    """MarkDownValidatorCrew crew"""
    agents_config = 'config/agents.yaml'
    tasks_config = 'config/tasks.yaml'

    @agent
    def RequirementsManager(self) -> Agent:
        return Agent(
            config=self.agents_config['Requirements_Manager'],
            tools=[markdown_validation_tool],
            allow_delegation=False,
            verbose=False,
            llm=LLM
        )

    @task
    def syntax_review_task(self) -> Task:
        return Task(
            config=self.tasks_config['syntax_review_task'],
            agent=self.RequirementsManager()
        )

    @crew
    def crew(self) -> Crew:
        """Creates the MarkDownValidatorCrew crew"""
        return Crew(
            agents=self.agents,
            tasks=self.tasks,
            process=Process.sequential,
            verbose=False,
        )

main.py

import sys

from markdown_validator.crew import MarkDownValidatorCrew


def run():
    """
    Run the markdown validation crew to analyze the markdown file.
    """
    # Get the input markdown file from command line arguments
    inputs = {
        'query': 'Please provide the markdown file to analyze:',
        'filename': sys.argv[1] if len(sys.argv) > 1 else None,  # Expect 'filename' key
    }

    # Check if the markdown file path is provided
    if inputs['filename']:
        print(f"Starting markdown validation for file: {inputs['filename']}")
        crewResult = MarkDownValidatorCrew().crew().kickoff(inputs=inputs)
        print("Markdown validation completed")
        return crewResult
    else:
        raise ValueError("Error: No markdown file provided. Please provide a file path as a command-line argument.")


def train():
    """
    Train the markdown validator crew for a given number of iterations.
    """
    # Get the number of iterations and markdown file path from command line arguments
    inputs = {
        'query': 'Training the markdown validation model.',
        'filename': sys.argv[2] if len(sys.argv) > 2 else None,  # Expect 'filename' key
    }

    # Check if the markdown file path is provided
    if inputs['filename']:
        try:
            print(f"Starting training for file: {inputs['filename']}")
            MarkDownValidatorCrew().crew().train(n_iterations=int(sys.argv[1]), filename=inputs['filename'])
            print("Training completed successfully.")
        except Exception as e1:
            raise Exception(f"An error occurred while training the crew: {e1}")
    else:
        raise ValueError(
            "Error: No markdown file provided for training. Please provide the number of iterations and a file path.")


if __name__ == "__main__":
    print("## Welcome to Markdown Validator Crew")
    print('-------------------------------------')

    try:
        result = run()
        print("\n\n########################")
        print("## Validation Report")
        print("########################\n")
        print(f"Final Recommendations: {result}")
    except Exception as e:
        print(f"An error occurred: {e}")

note

For more examples, check out the crewAI examples repository. Remember you will need to adapt them to use a custom Nuclia LLM.