JiMax Object Tool LLM

Generate structured objects using AI SDK with tools (run other n8n flows)

JiMax Object Tool LLM

AI-generated

Summary

Generate a structured object from a prompt with a selected AI provider, optionally calling configured n8n workflows as tools and using file-backed conversation memory.

Inputs

  • Provider and Model ID (required) — Choose OpenAI, Google Gemini, Anthropic, or an OpenAI-compatible provider, then specify the model ID. The matching provider credentials are required.
  • Prompt (required) — The user message sent to the model for this input item.
  • Zod Schema (required) — Zod expression defining the structure of the model's returned object; the variable z is available.
  • Tools — Optional tools that let the model call specified n8n workflows. Each tool has a required name, description, and workflow ID, plus a Zod input schema. When Include Parent Item Data is enabled, the current input item's JSON is passed as _parentData alongside the tool arguments.
  • Max Steps — Maximum number of tool-calling iterations; defaults to 5.
  • System Prompt — Optional system message included before the other messages.
  • Initial Messages (JSON) — Optional JSON array of starting messages. Each message must have a role of user, assistant, system, or tool and string content.
  • Use Memory — When enabled, loads and saves conversation history using the Session ID. History is stored in jimax_memory.json in the n8n process working directory.
  • Session ID (required) — Conversation key used when memory is enabled.
  • Memory Window — When memory is enabled, number of previous interactions to load; defaults to 5.
  • Save in Background — When memory is enabled, saves history without waiting when true; defaults to true.

Output shape

One output item per input item, each containing the model-generated object as its JSON data.

The returned object follows the configured Zod schema. Tool calls return the first output item's JSON from the selected workflow; if it returns no output, the tool reports a success message. With continue-on-fail enabled, an item that errors returns { "error": "..." }. Memory stores the latest 100 user/assistant messages per session.

Examples

Example 1: Use a model to classify a support request and call a workflow to look up customer details

Set a prompt and output schema, add a workflow tool with its input schema, and optionally enable memory with a session ID.

Links

Discussion