Agent Node

The Agent Node is a component in the Sequential Agents category, designed to execute tools and manage complex interactions within a conversational AI system. It can process inputs, utilize various tools, and generate responses based on predefined prompts and dynamic states.

Node Details

• Name: Agent_SeqAgents

• Type: Agent

• Category: Sequential Agents

• Version: 3.0

Input Parameters

1. Agent Name (string, required)

   • Label: Agent Name

   • Description: Unique identifier for the agent

2. System Prompt (string, optional)

   • Label: System Prompt

   • Default: Example prompt for a research assistant

   • Description: Initial instructions for the agent

3. Human Prompt (string, optional)

   • Label: Human Prompt

   • Description: Prompt added at the end of messages as a human message

4. Messages History (code, optional)

   • Label: Messages History

   • Description: List of messages between System Prompt and Human Prompt, useful for few-shot examples

5. Tools (Tool, list, optional)

   • Label: Tools

   • Description: List of tools available for the agent to use

6. Sequential Node (Start | Agent | Condition | LLMNode | ToolNode, list, required)

   • Label: Start | Agent | Condition | LLM | Tool Node

   • Description: Predecessor nodes in the sequential flow

7. Chat Model (BaseChatModel, optional)

   • Label: Chat Model

   • Description: Overwrite model to be used for this agent

8. Require Approval (boolean, optional)

   • Label: Require Approval

   • Description: If enabled, requires approval before executing tools

9. Format Prompt Values (json, optional)

   • Label: Format Prompt Values

   • Description: Assign values to prompt variables, can use state values

10. Approval Prompt (string, optional)

    • Label: Approval Prompt

    • Default: Default prompt asking for tool execution approval

    • Description: Custom prompt for approval requests

11. Approve Button Text (string, optional)

    • Label: Approve Button Text

    • Default: “Yes”

    • Description: Text for the approval button

12. Reject Button Text (string, optional)

    • Label: Reject Button Text

    • Default: “No”

    • Description: Text for the rejection button

13. Update State (tabs, optional)

    • Label: Update State

    • Description: Options to update the state after agent execution

    • Tabs: a. Update State (Table): UI-based state update b. Update State (Code): Code-based state update

14. Max Iterations (number, optional)

    • Label: Max Iterations

    • Description: Maximum number of iterations for the agent

Outputs

• output: The result of the agent’s execution, including:

  • Content: The main response text

  • Used Tools: List of tools used during execution

  • Source Documents: Any relevant source documents

  • Artifacts: Additional data or files generated

Functionality

1. Tool Execution: Can use various tools to accomplish tasks.

2. State Management: Updates and maintains state across interactions.

3. Approval Workflow: Optional approval process before tool execution.

4. Dynamic Prompting: Uses system and human prompts with variable substitution.

5. Multi-modal Support: Can process and generate responses with image content.

6. Iteration Control: Limits the number of execution cycles.

7. Flexible Configuration: Supports both UI-based and code-based configurations.

Use Cases

• Complex multi-step conversations

• Research and information gathering tasks

• Decision-making processes requiring multiple tools

• Workflows needing human oversight (with approval feature)

• Dynamic state-based interactions in chatbots or virtual assistants

Notes

• Requires proper setup of predecessor nodes and tools.

• Careful configuration of prompts and state management is crucial for effective operation.

• Can be computationally intensive, especially with multiple tools and iterations.