Skip to main content
HITL (Human In The Loop) is how software asks a person. You build a workflow that says who to ask, what to ask and what to do with the answer. Your agent, your code, n8n, Zapier or Make runs it. The people you picked get the question on their phone, in the web inbox or in Telegram, answer it, and the run finishes with their answer.
The HITL workflow editor showing a refund approval workflow: a webhook trigger, an Approve refund review step, an Approved? condition, two actions and an End step, with the review step's settings open on the right

​
How it works

1

Build a workflow

In the editor at my.hitl.sh, add a trigger and a review step: the question, the kind of answer you want (approve or deny, pick an option, a number, edited text, …) and a deadline. Add AI, condition and action steps around it if you like.
2

Invite the reviewers

Invite people by email, or share the workflow’s join code, QR code or link. They answer in the HITL app for iOS and Android, the web inbox or the Telegram mini app.
3

Run it

Start a run with the API, the CLI, the MCP server from Claude, Cursor or another agent, or from n8n, Zapier or Make. Each run is an execution.
4

Get the answer

The execution moves from running to pending_review while it waits for a person, then to completed with their decision. Read it with GET /executions/:id?wait=30, have it POSTed to a callback_url, or subscribe to execution.finished.

​
Start here

Quickstart

From nothing to a person approving your first run, in about five minutes.

Set up your agent

Paste one prompt into Claude Code, Codex or Cursor and let it connect itself.

Workflows

Triggers, steps, review steps and executions.

API reference

Run workflows and follow their runs over REST.

​
Ways to connect

MCP

Claude, Cursor and any MCP client

CLI

npx @hitl-sh/cli

REST API

Any language

n8n

Community node

Zapier

Run a Workflow action

Make

Verified app
Coming from HITL version 1? Read Moving from HITL version 1.