Databricks AI Agent icon

Databricks AI Agent

Databricks AI Agent with MLflow tracking. Generates an action plan and executes it. Can use external tools.

Databricks AI Agent

AI-generated

Overview

The Databricks AI Agent node integrates with Databricks AI to generate and execute an action plan using an AI agent. It supports input prompts either from a connected Chat Trigger node or defined manually. The node can optionally require a specific output format by connecting an output parser, enable MLflow tracking for logging execution traces, and use a fallback language model if the main model fails. It also supports various options such as system messages, maximum iterations, returning intermediate steps, automatic handling of binary images, batch processing for rate limiting, and streaming responses in real-time.

Use Case Examples

  1. Use the node to automate complex workflows by generating step-by-step action plans based on user prompts.
  2. Integrate with a chat interface where user messages are passed automatically to the agent for processing.
  3. Enable MLflow tracking to monitor and analyze the agent's execution performance over time.

Properties

Name Meaning
Source for Prompt (User Message) Determines the source of the user prompt, either from a connected Chat Trigger node or defined manually.
Prompt (User Message) The actual user message or prompt text to be processed by the AI agent.
Require Specific Output Format Boolean flag to indicate if a specific output format is required by connecting an output parser.
Enable MLflow Tracking Boolean flag to enable logging of agent execution traces to MLflow using the workflow ID as the experiment name.
Enable Fallback Model Boolean flag to enable use of an additional fallback language model if the main model fails.
Options Collection of additional configuration options including system message, max iterations, return intermediate steps, automatic passthrough of binary images, batch processing settings, and streaming enablement.

Output

JSON

  • result - The final output generated by the AI agent after processing the prompt and executing the action plan.
  • intermediateSteps - Optional output containing intermediate steps taken by the agent if enabled.
  • executionTrace - Optional MLflow execution trace logs if MLflow tracking is enabled.

Dependencies

  • Databricks API credentials (optional) for integration
  • MLflow for execution trace logging (optional)

Troubleshooting

  • Ensure the 'chatInput' field is correctly passed from the connected Chat Trigger node when using 'auto' prompt source.
  • If enabling MLflow tracking, verify that MLflow is properly configured and accessible.
  • When using fallback model, ensure the additional language model node is correctly connected and configured.
  • Streaming responses require the connected trigger node to be set to 'Streaming Response' mode to function properly.

Links

Discussion