Karini AI Documentation
Go Back to Karini AI
  • Introduction
  • Installation
  • Getting Started
  • Organization
  • User Management
    • User Invitations
    • Role Management
  • Model Hub
    • Embeddings Models
    • Large Language Models (LLMs)
  • Prompt Management
    • Prompt Templates
    • Create Prompt
    • Test Prompt
      • Test & Compare
      • Prompt Observability
      • Prompt Runs
    • Agentic Prompts
      • Create Agent Prompt
      • Test Agent Prompt
    • Prompt Task Types
    • Prompt Versions
  • Datasets
  • Recipes
    • QnA Recipe
      • Data Storage Connectors
      • Connector Credential Setup
      • Vector Stores
      • Create Recipe
      • Run Recipe
      • Test Recipe
      • Evaluate Recipe
      • Export Recipe
      • Recipe Runs
      • Recipe Actions
    • Agent Recipe
      • Agent Recipe Configuration
      • Set up Agentic Recipe
      • Test Agentic Recipe
      • Agentic Evaluation
    • Databricks Recipe
  • Copilots
  • Observability
  • Dashboard Overview
    • Statistical Overview
    • Cost & Usage Summary
      • Spend by LLM Endpoint
      • Spend by Generative AI Application
    • Model Endpoints & Datasets Distribution
    • Dataset Dashboard
    • Copilot Dashboard
    • Model Endpoints Dashboard
  • Catalog Schemas
    • Connectors
    • Catalog Schema Import and Publication Process
  • Prompt Optimization Experiments
    • Set up and execute experiment
    • Optimization Insights
  • Generative AI Workshop
    • Agentic RAG
    • Intelligent Document Processing
    • Generative BI Agentic Assistant
  • Release Notes
Powered by GitBook
On this page
  • Right Panel (Configuration Panel):
  • Download Descriptor:
  • Import Descriptor:
  • Number of State Updates:
  • Test Context:
  • Send Alerts:
  • Save and Test
  • Evaluation:
  1. Recipes
  2. Agent Recipe

Agent Recipe Configuration

PreviousAgent RecipeNextSet up Agentic Recipe

Last updated 1 month ago

To create a new agentic workflow recipe, go to the recipe page, click Add new, select the appropriate runtime option, provide a user-friendly name and detailed description, and choose Agent 2.0 for the recipe type.

You will observe the Left Panel, the blank canvas, and the Right Panel, which contains the following fields.

Right Panel (Configuration Panel):

The right panel of the interface provides several key fields that allow users to configure, test, and evaluate their workflow.

Download Descriptor:

This option allows the user to download the current recipe configuration in a descriptor format, typically in JSON The descriptor contains all the settings, structure, and logic of the workflow. It is an essential tool for:

  • Backing up: Storing the recipe's configuration for future use.

  • Sharing: Sharing the recipe configuration with others, enabling collaboration or deployment across different environments.

  • Version control: Tracking changes or maintaining multiple versions of the recipe.

By downloading the descriptor, users ensure that they have a portable version of the workflow that can be reused or referenced independently of the platform.

Import Descriptor:

This field allows users to import an existing descriptor from a file into the current workflow. Importing a descriptor is useful for:

  • Reusing existing recipes: Users can import pre-defined workflows, saving time by building upon an existing structure.

  • Version updates: If there are changes to a recipe that were saved previously in a descriptor file, users can import the latest version to update the workflow.

Number of State Updates:

This field specifies the number of state updates that the workflow will track or monitor during execution. A state update typically refers to a change in the workflow’s execution state, such as data transformations, decision points, or changes in the workflow’s logic. The field:

  • Controls monitoring: Defines how many times the state of the recipe will be tracked, helping users to optimize performance.

  • Regulates execution tracking: Useful for complex workflows, where tracking multiple state changes helps in debugging or performance tuning.

For example, if the workflow is expected to go through 10 steps, the number of state updates could be set to 10 to track each transition between states.

Test Context:

The Test Context field allows users to input sample data or parameters to simulate and test the recipe. This provides a way for users to:

  • Validate functionality: By testing the workflow with actual or mock data, users can ensure that the recipe performs as expected before going live.

  • Debugging: Users can detect potential issues by testing with different inputs and observing the output.

  • Customization: It allows the user to simulate various contexts (such as different user inputs or external data) to see how the workflow responds under different conditions.

In essence, this field is for testing and debugging the recipe to ensure its functionality and proper behavior with real or mock data.

Send Alerts:

The Send Alerts feature enables users to configure automated notifications for specific events or errors within the system. This is especially useful for real-time monitoring and proactive issue resolution. This feature is primarily useful for:

  • Error Handling: Notifying the user if the workflow encounters an issue or failure, enabling prompt intervention.

  • Proactive Notifications: Ensuring users are automatically informed of critical issues without the need for continuous manual monitoring.

The following image illustrates the detailed sections.

Recipient Emails:

This section allows users to input email addresses that will receive the alert notifications. By pressing Enter or Tab, users can add multiple recipient emails. This enables the distribution of alerts to the necessary stakeholders or support teams.

Subject: The subject line of the alert email is automatically populated or can be customized based on the event. It provides a quick summary of the alert, such as a specific issue or status update.

Message: The message body is structured to include dynamic placeholders such as:

  • {metadata.datetime}: This placeholder will be replaced with the current date and time when the alert is generated.

  • {metadata.message_url}: This dynamic link will direct the recipient to the relevant error or message URL for further details about the issue.

The message also provides details on the specific error that occurred, helping recipients quickly understand the context and take appropriate action.

Message template:

Hello,

Date: {metadata.datetime}

An error occurred in your recipe: {metadata.message_url}.

Error Details:
{errorMessage}

Please review the error message and address the issue at your earliest convenience.

Thank you!

Save and Test

This button serves two primary functions:

  1. Save: It saves the current configuration of the recipe, ensuring that any changes made up to this point are preserved.

  2. Test: The Test functionality initiates a test execution of the recipe, simulating the workflow’s operation based on the provided inputs and test context. To enable this functionality, a Chat node must be included in the recipe. Upon clicking the Test button, the system will populate the chat window, allowing the user to input a question and evaluate the response generated by the workflow.

This feature is significant for:

  • Validating the logic: Ensuring that the workflow executes as expected.

  • Identifying issues early: Catching errors or unexpected behaviors before the recipe goes live.

  • Confidence before deployment: Testing allows users to confirm the recipe’s functionality and troubleshoot before full-scale execution.

This button is integral to refining and testing the workflow in a controlled environment before it is put into production.

Evaluation:

The Evaluation section allows users to assess the performance of the recipe by uploading a dataset. Users can choose between two evaluation options:

  • Upload Dataset: This option allows the user to upload real or test data to evaluate how well the recipe performs with different inputs. It is a way to test the recipe with diverse data scenarios.

  • Default Evaluation: This applies the platform's pre-defined evaluation metrics to automatically assess the recipe’s performance based on general or default criteria.

  • Custom Evaluation: Allows the user to define specific evaluation criteria tailored to their use case. This flexibility is useful when users have unique performance metrics or requirements for how the recipe should behave.

Evaluation helps ensure that the recipe is optimized, efficient, and performs as expected under various conditions, particularly with real-world data. For further information, please refer to section.

Evaluation