HITL on latest 2.10.4

Describe the problem/error/question

When using the Human-in-the-Loop (HITL) tool nodes with an AI Agent + Chat Trigger, approving the request via an external channel (Gmail, SMTP) does not return the tool result back to the AI Agent. The tool fires correctly after approval and the execution log confirms the downstream tool (Google Sheets) runs successfully — but the AI Agent never receives the result and the chat window responds with “no item to return was found”.

The workflow works perfectly when the HITL node is removed — the Google Sheets tool fires and the result returns to the AI Agent and the chat interface as expected. The issue is specifically that after the async email approval resumes the execution, the tool result is not passed back up to the AI Agent’s reasoning loop.

I tested with both Gmail HITL (gmailHitlTool, typeVersion 2.2) and an SMTP-based approval node — same behavior on both. The approval channel makes no difference.

This appears related to GitHub Issue #25491 (“Chat Hub currently not supporting Human Review UI”), though that issue focused on the Chat approval channel. The underlying problem — HITL tool results not propagating back to the AI Agent after async approval — appears to affect all external channels (Gmail, SMTP confirmed).


What is the error message (if any)?

No explicit error. The AI Agent responds in chat with:

“no item to return was found”

The execution log shows the Google Sheets tool node ran successfully after approval, but the result never surfaces in the chat response.


Please share your workflow

json


Share the output returned by the last node

The Google Sheets node returns one row of data successfully (visible in the execution log). However, after approval resumes the workflow, this result never propagates back to the AI Agent. The chat interface receives no data and the agent responds with “no item to return was found.”

When HITL is removed and the Google Sheets tool connects directly to the AI Agent, it works correctly — data is returned and displayed in chat as expected.


Information on your n8n setup

  • n8n version: 2.10.4
  • Database (default: SQLite): SQLite (default)
  • n8n EXECUTIONS_PROCESS setting (default: own, main): Not applicable — running with External Task Runners (N8N_RUNNERS_MODE=external)
  • Running n8n via: Docker (self-hosted), two containers — n8n (main app) and n8n-runners (external task runner for JavaScript/Python execution)
  • Operating system: macOS (latest) / Docker Desktop (latest)

Related GitHub issue: #25491 — Chat Hub currently not supporting Human Review UI (closed Feb 8, 2026 — fix confirmed in dev build for Chat channel, but external channel HITL tool result propagation appears still broken)

Yeah this is a known bug, the async resumption from email approval doesn’t feed the result back into the agent’s tool loop. There are a few open GitHub issues tracking it (#22483, #26373). No real workaround right now other than removing the HITL node unfortunately, which you already figured out. Worth commenting on those issues so the team knows more people are hitting it.

I’ll wait for the fix as I am recording a video update for my course on Udemy, and I try to announce new features. This looks bad announcing something that both new and broken and I don’t want to turn off folks that are new to n8n.

hi @GenXerator

The most reliable pattern I’ve seen in this case is replacing the HITL node with a Wait + Webhook approval flow. Instead of pausing inside the agent loop, the workflow sends an approval email containing a webhook link, pauses with a Wait node, and resumes when the webhook is called. After that the tool can run and the result can be returned normally.

Another option is splitting the process into two workflows, one that creates the approval request and another that handles the approval and executes the tool. This avoids relying on the agent context surviving an async pause.

As as 2.13.2 this newly announced feature appears to still not be working as expected. It’s slightly better in that after the email approval is received, it does execute the sheet tool to retrieve the data after HITL approval is given… but now as of 2.13.2 that data is not passed back into the AI Agent to return any creditable response. the AI Agent seams ignorant of the tools execution.

I don’t understand how a new feature can exist for so many releases and still be broken. Feels like QA ias broken or the MTC’s are missing from the QA process. Again my interest is for my students of my n8n course on Udemy, so I’m not looking for workarounds as I would be for a production use.

Usually the folks at n8n are pretty snappy at addressing issues. What I do not get his is a highly visible feature as its still showing as a highlighted feature in the release notes.

Removing the HITL feature, the chatflow works as expected. With HITL, I expect after the HITL approval that the same or similar output shows. IME/IMO this extremely stupid and simple use case, this feature appears to STILL be broken as of 2.13.2

Hi @GenXerator

Please see if the new version meets your needs.
Release [email protected] · n8n-io/n8n

  1. it’s pre-release and I would not announce a feature thats still buggy and not fixed in an actual release - less student questions flood my inbox asking what is broken or what they are doing wrong or why cant they get it to work - its a support nightmare for me so I only release videos on features I can demo and confirm working.

  2. The release provides no fixes at 1st glance

  3. Partial execution of Chat node and Chat tool (#26334) (a7b9e00) looked promising but was not CLEAR this fixes it fix: Partial execution of Chat node and Chat tool by michael-radency · Pull Request #26334 · n8n-io/n8n · GitHub

  4. I installed it anyways to try and NO SOSDD.