Action Types

Agentic AI supports multiple action types, each designed for a specific use case.

Answer from Web

Retrieves real-time information by searching the web.

Use when:

Answers require up-to-date information from the internet.

Key Points

Does not use the Agent's System Prompt.

Cannot be restricted to specific domains through prompts or system instructions.

The Agent does not have visibility into the action's prompt, execution, or response.

Code Node

Executes custom Python logic to perform backend business operations and integrate with external systems.

Use when:

Making API calls to external systems

Executing custom Python logic

Validating parameters against complex business rules

Performing computations based on captured parameters

Processing data or interacting with databases

Key Points

Enables execution of custom Python code for advanced business logic.

Can be used to call external APIs, validate complex conditions, transform data, and perform calculations.

Provides the Agent with full execution context, making it ideal for workflows that require dynamic processing or external integrations.

Catalog Search

Searches product catalogs to provide relevant product recommendations during commerce journeys.

Use when:

Building e-commerce or shopping experiences.

Recommending products based on user preferences or search criteria.

Prompt

Executes a dedicated prompt for specific scenarios without relying on the Agent's main instructions.

Use when:

Breaking large instruction sets into smaller, reusable modules.

Managing static FAQs or specialized knowledge.

Reducing the size of the main System Prompt.

Key Points

Uses only the prompt configured for the action.

Does not inherit the Agent's System Prompt.

Recommended for prompts under 40,000 tokens.

Retrieval-Augmented Generation (RAG)

Retrieves information from a predefined set of documents or a restricted knowledge base to generate accurate, domain-specific responses.

Use when:

Answering questions from internal documentation, policies, manuals, or knowledge bases.

Key Points

Uses only the prompt configured for the RAG action.

Does not inherit the Agent's System Prompt or conversation context.

Processes only the queries passed to the RAG action.

For multi-turn conversations requiring full context, continue the workflow within the Agent after the RAG response.

RAG prompts are limited to 1,000 characters.