Skip to main content

LLM Agent

An intelligent agent powered by Large Language Models (LLMs) that can process queries, use tools, manage sub-agents, and maintain conversation state across sessions.

Common Properties​

  • Name - The custom name of the node.
  • Color - The custom color of the node.
  • Delay Before (sec) - Waits in seconds before executing the node.
  • Delay After (sec) - Waits in seconds after executing node.
  • Continue On Error - Automation will continue regardless of any error. The default value is false.

Inputs​

  • Session ID (optional) - (string) A unique identifier for this specific conversation thread, essential for retrieving it later.
  • Name - (string) Name of the agent.
  • Description - (string) Description of the agent's purpose and capabilities.
  • LLM Agent Instructions - (string) Instructions for the agent. Supports Mustache templates with session state and artifact access. Use {{key}} for variables, {{#key}}...{{/key}} for conditionals/loops, {{#artifacts}}...{{/artifacts}} to iterate over available files, and {{file:filename}} for file references.
  • Query - (string) User query to the agent. Supports Mustache templates with full artifact embedding. Use {{file:filename}} to include actual file content in the query.
  • Files - (string) Files from Artifact Service to load and send to agent. Can be JSON array with filename/version: [{"filename": "image.jpg", "version": 1}] or string array: ["image.jpg", "document.pdf"]. Version is optional (defaults to latest).
  • Output Key - (string) Automatically saves the agent's final response text (or structured output) to the specified state key.

Options​

  • Model Name - LLM model to use. Options include:
    • Google Models: Gemini 3 Pro, Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash Lite, Gemini 2.0 Flash
    • OpenAI Models: GPT-5, GPT-5 mini, GPT-5 nano, o3, o4-mini, GPT-4o, GPT-4o mini
    • Anthropic Models: Claude Opus 4.5, Claude Sonnet 4.5, Claude Sonnet 4, Claude Opus 4
    • Custom Model (specify manually)
  • Custom Model - (string) Enter the specific model string if 'Custom Model' was selected. Examples: groq/llama3-8b-8192, ollama/mistral, gemini-experimental.
  • API Key - (Credentials) API Key if required for direct provider SDKs and API access.
  • Built-in Tool - Include a built-in tool with the agent. Options: None, Google Search, Code Execution.
  • API Base URL (Optional) - (string) Specify the API base URL for vLLM or self-hosted servers (e.g., http://localhost:8000/v1 for vLLM, http://localhost:11434 for Ollama).
  • Output Schema (JSON) - (string) Optional JSON schema to enforce structured output format. Note: Using output_schema prevents the agent from using tools.
  • Timeout (seconds) - (number) Maximum wait time in seconds for agent operations to finish. Default: 300.
  • Session Service - Session storage backend for conversation context. Options: In Memory Session Service, Robomotion Session Service.
  • Artifact Service - Storage backend for files and artifacts. Options: In Memory Artifact Service, Robomotion Artifact Service.

Outputs​

  • Text - (string) Agent text response.
  • Images - (array) Images returned by the agent saved as temp files (array of file paths).
  • File URIs - (array) File URI references returned by the agent (array of URI strings like gs://...).
  • Code Outputs - (array) Code execution output strings (array of stdout strings).
  • Raw Response - (object) Full API response structure (only set when Include Raw Response is enabled).

Ports​

  • sub-agents - (output) Connect to other agents (LLM Agent, Sequential Agent, Parallel Agent, Loop Agent) to create hierarchical agent systems.
  • tools - (output) Connect to Tool In nodes to give the agent custom capabilities.
  • callbacks - (output) Connect to Callback In nodes to monitor and intercept agent operations.

How It Works​

The LLM Agent node creates an AI agent that can:

  1. Process Natural Language - Accepts queries and responds using the configured LLM model
  2. Maintain Context - Uses session management to remember conversation history
  3. Use Tools - Can call connected tools to perform actions (web search, code execution, custom workflows)
  4. Manage Sub-Agents - Delegates tasks to specialized sub-agents for complex workflows
  5. Handle Files - Processes images, PDFs, and other files through the artifact service
  6. Execute Code - When Code Execution is enabled, can write and run Python code
  7. Structured Output - Can return responses in a specified JSON schema format

The agent uses the Google Agent Development Kit (ADK) internally and supports multimodal interactions with text, images, and files.

Common Use Cases​

  • Conversational AI - Build chatbots and virtual assistants
  • Document Analysis - Process and extract information from PDFs, images, and text files
  • Multi-Step Workflows - Combine agents and tools to automate complex tasks
  • Code Generation - Generate and execute code to solve problems
  • Web Research - Search and synthesize information from the web
  • Image Analysis - Analyze and describe images using vision models

Error Handling​

The node handles various error scenarios:

  • Invalid API credentials
  • Model not available or unsupported
  • Timeout during agent execution
  • Tool execution failures
  • Session or artifact service connection issues

When errors occur, the agent will attempt to provide a user-friendly error message and continue execution if Continue On Error is enabled.

Usage Notes​

  • Session Management - The Session ID input allows you to continue previous conversations. If not provided, a new session will be created automatically.
  • Template Support - Both Instructions and Query support Mustache templates, allowing dynamic content based on session state.
  • File Handling - Files can be referenced in queries using {{file:filename}} syntax to embed their content.
  • Tool Limitations - When using Output Schema for structured output, tools cannot be used simultaneously.
  • Model Selection - Different models have different capabilities, costs, and performance characteristics. Choose based on your needs.
  • Robomotion Credits - When running in Robomotion cloud, you can use the managed AI service without providing API keys.
  • Code Execution - The built-in code executor runs Python code in a sandboxed environment with common libraries available.
  • Callbacks - Use callbacks to monitor agent behavior, log interactions, or implement custom logic at various stages.