AI Activities

Use the AI group to build workflows that call language models, generate embeddings, store and retrieve vector data, maintain conversational memory, invoke MCP tools, or coordinate an AI agent.

Available activities

Activity	Purpose
LLM Provider	Defines and calls a supported language-model provider
Embedding Model	Converts text or other supported content into vector embeddings
Vector DB	Stores and searches vector data
Memory	Provides durable or provider-backed conversational memory
Simple Memory	Provides lightweight organization-level memory for straightforward flows
MCP	Makes configured Model Context Protocol tools available to an AI flow
AI Agent	Coordinates a model, instructions, memory, and tools to complete a task

Build an AI workflow

1. Configure the required provider and credentials as resources or managed secrets.
2. Add LLM Provider for direct model calls, or AI Agent for tool-using, multi-step behavior.
3. Add memory only when the workflow needs context across interactions.
4. Add Embedding Model and Vector DB for retrieval-augmented generation or semantic search.
5. Add MCP only for approved servers and tools.
6. Map model output through validation and business rules before it reaches a write operation.

Safety and governance

• Do not place secrets, regulated data, or unnecessary personal information in prompts.
• Treat model output as untrusted input and validate it before executing commands or changing records.
• Restrict MCP and other tools to the minimum required permissions.
• Require human confirmation for consequential or irreversible actions.
• Set token, timeout, and iteration limits to control cost and prevent loops.
• Record provider, model, prompt version, tool calls, and outcome where audit policy permits.

For AI-assisted application development in Studio, see the NxAgent user guide. For publishing API operations as AI tools, see MCP Servers.