Agent Kit

Agente de IA com carregamento dinâmico de skills, memória persistente e suporte a ferramentas MCP.

Agent Kit

AI-generated

Summary

Process input messages through a configurable AI agent that dynamically loads skills, manages session-based memory, and optionally uses external tools, returning the agent's response with content guardrails applied pre- and post-processing.

Inputs

  • Campo da Mensagem de Entrada (required) — JSON field name containing the user message to process.
  • Campo do ID de Sessão (required) — JSON field name used to identify the session for persistent memory.
  • System Prompt Base (required) — Base system prompt defining agent behavior and personality.
  • Modelo Alternativo — Overrides the default language model specified in credentials.
  • Máximo de Iterações (required) — Maximum number of agent iterations when invoking tools before returning a response.
  • Campo de Saída (required) — JSON field name to store the agent's response output.
  • Campo de Skills (required) — JSON input field path to load dynamic skills (supports dot/bracket notation).
  • Skills Inline — Skills defined directly in the node; combined with loader skills, with loader taking priority.
  • Guardrails — Content guardrails evaluated before (pre) and after (post) agent processing to block or filter input/output.

Output shape

a list of JSON objects (one per input item) each containing the original data minus internal skill fields plus the agent's response under the specified output field. Includes usage metrics and model info.

If guardrails block the input or output content, the response field contains the guardrail's fallback message. Supports optional persistent memory and MCP tools passed in connected input nodes. Throws errors if the input message is missing or agent fails to produce a response after max iterations.

Links

Discussion