hitlbeginner10 min setupby Mantle

AI screen, human approve, then send

Classify an inbound request, ask a reviewer to approve or reject it, and email the requester the outcome.

Request this template in a demo Talk to sales

Drag to pan, pinch to zoom. This is the real graph the template installs.

Purpose

Classify an inbound request, ask a reviewer to approve or reject it, and email the requester the outcome.

How it works

What it does

Webhook approval-request → Risk screen (low_risk | needs_review | high_risk) gives the reviewer context.
Reviewer decision pauses the run until someone picks Approve / Reject (48 h timeout).
Approved? routes to the approval or rejection email.

Nodes 6

01Request intrigger.webhookTrigger
02Risk screenai.classifyAI
03Reviewer decisionhuman.promptHuman
04Approved?control.branchLogic
05Send approvalmail.sendDestination
06Send rejectionmail.sendDestination

View template JSON

{
  "format": "mantle-workflow/v1",
  "id": "hitl-approval-gate",
  "name": "AI screen, human approve, then send",
  "graph": {
    "nodes": [
      {
        "id": "webhook",
        "type": "trigger.webhook",
        "label": "Request in",
        "config": {
          "slug": "approval-request",
          "input_schema": {
            "required": [
              "request",
              "requester_email"
            ],
            "properties": {
              "request": {
                "type": "string"
              },
              "requester_email": {
                "type": "string"
              }
            }
          }
        },
        "worker_type": "generic-worker",
        "position": {
          "x": 60,
          "y": 220
        }
      },
      {
        "id": "ai_classify",
        "type": "ai.classify",
        "label": "Risk screen",
        "config": {
          "model": "claude-haiku-4-5",
          "labels": [
            "low_risk",
            "needs_review",
            "high_risk"
          ],
          "text": "{{ $json.request }}",
          "temperature": 0
        },
        "worker_type": "ai-worker",
        "position": {
          "x": 380,
          "y": 220
        }
      },
      {
        "id": "reviewer",
        "type": "human.prompt",
        "label": "Reviewer decision",
        "config": {
          "prompt_text": "Request from {{ steps.webhook.requester_email }} (AI: {{ steps.ai_classify.response.label }}, {{ steps.ai_classify.response.confidence }}):\n\n{{ steps.webhook.request }}",
          "options": [
            {
              "id": "approve",
              "label": "Approve"
            },
            {
              "id": "reject",
              "label": "Reject"
            }
          ],
          "timeout_seconds": 172800
        },
        "worker_type": "generic-worker",
        "position": {
          "x": 700,
          "y": 220
        }
      },
      {
        "id": "branch_approved",
        "type": "control.branch",
        "label": "Approved?",
        "config": {
          "mode": "first_match"
        },
        "worker_type": "generic-worker",
        "position": {
          "x": 1020,
          "y": 220
        }
      },
      {
        "id": "send_approved",
        "type": "mail.send",
        "label": "Send approval",
        "config": {
          "to": "{{ steps.webhook.requester_email }}",
          "subject": "Your request was approved",
          "body_text": "Good news — your request has been approved.\n\n{{ steps.webhook.request }}"
        },
        "worker_type": "generic-worker",
        "position": {
          "x": 1340,
          "y": 30
        }
      },
      {
        "id": "send_rejected",
        "type": "mail.send",
        "label": "Send rejection",
        "config": {
          "to": "{{ steps.webhook.requester_email }}",
          "subject": "Your request was not approved",
          "body_text": "Unfortunately your request was not approved. Reply to this email if you have questions.\n\n{{ steps.webhook.request }}"
        },
        "worker_type": "generic-worker",
        "position": {
          "x": 1340,
          "y": 410
        }
      }
    ],
    "edges": [
      {
        "from": "webhook",
        "to": "ai_classify"
      },
      {
        "from": "ai_classify",
        "to": "reviewer"
      },
      {
        "from": "reviewer",
        "to": "branch_approved"
      },
      {
        "from": "branch_approved",
        "to": "send_approved",
        "branch": "true",
        "condition": {
          "match": "all",
          "rules": [
            {
              "left": "response",
              "op": "equals",
              "right": "approve"
            }
          ]
        },
        "label": "approve"
      },
      {
        "from": "branch_approved",
        "to": "send_rejected",
        "branch": "false",
        "label": "reject",
        "priority": 1
      }
    ]
  }
}

Requirements

The connections and permissions each step needs, from the node contracts in the catalog.

AnthropicAPI key
Used by Risk screen.
Credential: {"api_key"}
- The provider adapter is chosen from the connection; `model` must belong to it.
SendGridAPI key
Used by Send approval, Send rejection.
Credential: {"api_key"} or {"username", "password"}

Setup

AI connection on Risk screen; SendGrid/SMTP on both send nodes.
Optional: add a second branch before the reviewer to auto-approve low_risk requests.

Sample input and output

Synthetic examples: node-contract samples or placeholders, never customer data.

Input

The fields the template declares or reads from the request body. Values are placeholders.

{
  "request": "<request>",
  "requester_email": "<requester_email>"
}

Output

Send approval mail.send

{
  "body": null,
  "message_id": "",
  "provider": "",
  "sent": true,
  "status_code": 0,
  "thread_id": ""
}

Send rejection mail.send

{
  "body": null,
  "message_id": "",
  "provider": "",
  "sent": true,
  "status_code": 0,
  "thread_id": ""
}

Failure handling

What the graph does when a step fails, read from its retry settings, error branches and error policies.

Reviewer decision: the step fails if nobody answers within 2 days.
A step that fails, after any retries, fails the run. The failed step is kept with its input and error.