Describe the problem/error/question
I’m building a conversational Orchestrator (GPT-4.1/4o-mini) that should silently delegate to four AI Agent Tool sub-agents (Search Setup, Detail Filters & Selection, CRM, Presentation).
Conversation quality is OK, but the orchestrator does not call the sub-agents. With Return Intermediate Steps enabled I don’t see tool calls; it just chats back to the user that is going to do something but doesn’t execute anything.
What I’ve already tried (and what changed):
-
Tool names: Renamed every tool/node to match
^[a-zA-Z0-9_-]+$(e.g.,BlockA_SearchSetup,BlockB_DetailFilters_Selection, etc.). This fixed the initial validation error. -
Arguments: In each AI Agent Tool prompt I inject inputs using
$fromAI(...)so the tool receivesuser_message/state(and optionallyuser_selection_ids). -
Orchestrator prompt: Updated to explicitly say it MAY CALL those tools silently and to pass the exact argument names.
-
Despite this, delegation still doesn’t happen. If there’s up-to-date guidance or best-practice prompt patterns for the AI Agent Tool and Orchestrator combo, I’d really appreciate it.
Please share your workflow
Share the output returned by the last node
-
From a cold start with input “hola” the Orchestrator replies in Spanish (greeting + question), but no tool calls show up in Intermediate Steps.
-
When asking “Quiero comprar un departamento en Coquimbo”, same behavior: chat reply only; still no delegation.
(Expected behavior: Orchestrator silently calls BlockA_SearchSetup(user_message, state); the tool receives those via $fromAI, returns JSON to its Structured Output Parser; then the Orchestrator continues.)
Information on your n8n setup
- n8n version: 1.107.3
- Database (default: SQLite):
- n8n EXECUTIONS_PROCESS setting (default: own, main):
- Running n8n via (Docker, npm, n8n cloud, desktop app): desktop app
- Operating system: