Agents

Agent

The Agent node represents a significant advancement in Lleverage's capabilities, allowing you to create autonomous AI agents that can independently pursue goals using specified capabilities.

Understanding Agents

An agent is an AI-powered component that can make its own decisions about how to accomplish a goal. Unlike standard nodes that follow predefined paths, an agent can:

• Determine which actions to take

• Use assigned capabilities in the optimal sequence

• React to intermediate results

• Persist until the goal is achieved

Agent Configuration

When adding an Agent node to your workflow, you'll configure several key elements:

Model Selection

Choose the AI model that will power your agent. Different models offer varying levels of reasoning ability, with options like:

• GPT-4o Mini

• Claude 3.7 Sonnet

• Other supported models

The model's capabilities directly impact how effectively your agent can reason about complex tasks and goals.

Goal Definition

The goal is what you want your agent to accomplish. This should be a clear, specific instruction that gives the agent direction without prescribing exactly how to achieve it. Goals can:

• Be entered directly in the Goal field

• Reference variables from previous nodes using the {{variableName}} syntax

• Include multiple objectives that are part of the same overall task

Examples:

• "Analyze this customer feedback and categorize the issues mentioned"

• "Research the top competitors for {{companyName}} and summarize their strengths"

• "Extract key performance metrics from this report and create a summary table"

Capabilities Assignment

Capabilities are the tools your agent can use to accomplish its goal. Each capability represents an integration or action the agent can perform when needed. The agent will decide which capabilities to use and in what order based on its understanding of the goal.

Available capabilities include:

• Searching the web

• Searching a knowledgebase

• Extracting web pages

• Using integrations

• And many other tools available in Lleverage

You can add multiple capabilities to give your agent a broader toolkit for solving problems.

How Agents Work

When an agent node runs:

1. The model receives the goal and information about available capabilities

2. It formulates a plan to achieve the goal using the capabilities

3. The agent executes capabilities as needed, processing the results at each step

4. It continues working until the goal is achieved or no further progress can be made

5. The final result is returned as the agent's output

Agent Output

The agent produces a comprehensive output that includes:

• The final result (the answer to the goal)

• A trace of its reasoning process

• A record of which capabilities were used and why

You can access this output in subsequent nodes using {{agent.output}}.

Best Practices for Agents

To get the most from agent nodes:

• Define clear goals: The more specific and unambiguous your goal, the better results you'll get

• Provide relevant capabilities: Only give the agent capabilities it needs for the task

• Start simple: Begin with straightforward goals before moving to complex multi-step tasks

• Review reasoning: Examine how the agent approached the problem to improve future workflows

• Consider model selection: More powerful models may deliver better results for complex reasoning tasks

Example Use Cases

Agents excel at tasks requiring judgment and multiple steps:

• Research automation: "Research the latest developments in quantum computing and summarize key breakthroughs"

• Data analysis: "Analyze these customer survey results and identify the top three areas for improvement"

• Content creation: "Create a blog outline about sustainable practices in manufacturing based on our latest report"

• Process automation: "Track down all stakeholders for project ABC-123 and summarize their latest updates"