Team Agents

Team Agents are a sophisticated type of agent on the aiXplain platform, designed for handling complex, multi-step tasks that require coordination between multiple components. By leveraging a system of specialized agents, tools, and workflows, Team Agents excel in scenarios that demand advanced task planning, quality control, and adaptability.

Team Agent Architecture

Mentalist: Handles task decomposition, planning, and dynamic re-planning as needed.
Inspector: Ensures outputs meet user-defined criteria, focusing on quality and consistency.
Bodyguard: (Coming Soon) Implements security measures, enforcing data access and privacy policies.
Orchestrator: Central LLM that coordinates tasks and invokes appropriate tools or agents.

Example

This example demonstrates a team agent that combines the functionalities of individual agents for text analysis, multimedia processing, and other tasks. It uses a specified large language model (LLM) to coordinate the agents.

from aixplain.factories import AgentFactory
from aixplain.modules.agent.tool.model_tool import ModelTool
from aixplain.factories import TeamAgentFactory

text_analysis_agent = AgentFactory.create(
    name="Text Analysis Agent",
    description="An AI assistant that analyzes text for sentiment.",
    instructions="""
    For a given input from the user, provide a sentiment analysis of the provided text and return the sentiment back to the user.
    """,
    tools=[
        AgentFactory.create_model_tool("615f4c626eb56373643b09d3")  
    ],
)

multimedia_agent = AgentFactory.create(
    name="Multimedia Agent",
    description="An AI assistant that takes in audio input and extract text from it.",
    instructions="""
    For a given input from the user, extract text from the audio input and return the text back to the user.
    """,
    tools=[
        AgentFactory.create_model_tool("60ddefab8d38c51c5885ee38")
    ]
)

team = TeamAgentFactory.create(
    name="Team of Agents for Text Audio and Image Processing",
    description="A team of agents that can process text and audio data.",
    agents=[
        text_analysis_agent,
        multimedia_agent,
    ],
    llm_id="6646261c6eb563165658bbb1"
)

response = team.run("Give me the sentiment from this audio ", content=['Audio File.mp3'])
response['data']['output']

Show output

Guide to Building the Example Team Agent

Create the Agents

Each specialized agent is designed with a specific role, including a name, description, instructions, and tools.

Text Anlaysis Agent

The Text Analysis Agent analyzes sentiment from a given text input using the Sentiment Analysis model by Google.

from aixplain.factories import AgentFactory
from aixplain.modules.agent.tool.model_tool import ModelTool

text_analysis_agent = AgentFactory.create(
    name="Text Analysis Agent",
    description="An AI assistant that analyzes text for sentiment.",
    instructions="""
    For a given input from the user, provide a sentiment analysis of the provided text and return the sentiment back to the user.
    """,
    tools=[
        AgentFactory.create_model_tool("615f4c626eb56373643b09d3") 
    ],
)

Multimedia Agent

The Multimedia Agent processes audio input and extracts text from speech. It uses this Speech Recognition model by Azure.

multimedia_agent = AgentFactory.create(
    name="Multimedia Agent",
    description="An AI assistant that takes in audio input and extract text from it.",
    instructions="""
    For a given input from the user, extract text from the audio input and return the text back to the user.
    """,
    tools=[
        AgentFactory.create_model_tool("60ddefab8d38c51c5885ee38")
    ]
)

Initialize the Team Agent

The Team Agent combines multiple specialized agents to process text and audio data. It is configured with a name, description, a list of specialized agents, and OpenAI's GPT 4o as the LLM to manage interactions.

from aixplain.factories import TeamAgentFactory

team = TeamAgentFactory.create(
    name="Team of Agents for Text Audio and Image Processing",
    description="A team of agents that can process text and audio data.",
    agents=[
        text_analysis_agent,
        multimedia_agent,
    ],
    llm_id="6646261c6eb563165658bbb1"
)

Run the Team Agent

Execute the team agent with specific inputs and evaluate its combined response.

response = team.run("Give me the sentiment from this audio ", content=['Audio File.mp3'])
response['data']['output']

Update the Team Agent

Make modifications to the team agent configuration or replace agents as needed. Use the save method to save changes.

new_agent = AgentFactory.create(
    name="Search Agent",
    description="Conduct search based on given topic."
)
new_agent

team.agents.append(new_agent)
team.save()

Debug the Team Agent

Validate individual agent performance and outputs.
Assess agent interactions for seamless execution.
Fine-tune the LLM and agents for better optimization.

Use data.intermediate_steps to analyze task execution and intermediate outputs:

response['data']['intermediate_steps']

Show output

Deploy the Team Agent

The team agent is initially stored as a draft, which remains available for 24 hours before being deprecated unless deployed. Deploying the team agent saves its configuration to the platform, making it available for production use and assigning an API endpoint for integration.

team.deploy()

Team Agent Parameters

Defines the configuration options for creating a team agent, including its name, included agents, LLM, and advanced features.

Parameter	Type	Default	Description
`name`	`Text`	[required]	The unique name of the team agent.
`agents`	`List[Text \| Agent]`	[required]	A list of agent names or objects to include in the team agent.
`llm_id`	`Text`	`"669a63646eb56306647e1091"`	ID of the large language model (LLM) powering the team agent.
`description`	`Text`	`""`	A brief description of the team agent's purpose.
`api_key`	`Text`	`config.TEAM_API_KEY`	API key for authenticating requests to manage the team agent.
`supplier`	`Dict \| Text \| Any \| int`	`"aiXplain"`	The owner or provider of the team agent.
`version`	`Text \| None`	`None`	The version number of the team agent for tracking updates.
`use_inspector`	`bool`	`True`	Enables an internal debugging and logging tool to track agent execution and performance.
`use_mentalist_and_inspector`	`bool`	`True`	Enables advanced features for supervision and planning when a supervisor is used.

Team Agent Methods

Provides key functions to create, list, and retrieve team agents on the platform.

Method	Description
`create`	Creates a new Team Agent with specified parameters, agents, and configurations.
`list`	Lists all Team Agents available on the platform for the given API key.
`get`	Retrieves the details of a Team Agent using its unique ID.
`run`	Executes the Team Agent on specified input data or query, providing results.
`run_async`	Executes the Team Agent asynchronously, returning a polling URL for the response.
`delete`	Deletes the Team Agent configuration from the platform.
`to_dict`	Converts the Team Agent's configuration and metadata to a dictionary format.
`validate`	Validates the Team Agent by checking its agents, name, and associated LLM.
`update`	Updates the Team Agent with new configurations and metadata.
`deploy`	Deploys the Team Agent, changing its status from `DRAFT` to `ONBOARDED`.

Create a team agent by following this multi-purpose text agent tutorial.

Team Agent Architecture​

Example​

Guide to Building the Example Team Agent​

Create the Agents​

Text Anlaysis Agent​

Multimedia Agent​

Initialize the Team Agent​

Run the Team Agent​

Update the Team Agent​

Debug the Team Agent​

Deploy the Team Agent​

Team Agent Parameters​

Team Agent Methods​