The Inbox Router

Description

GOAL
Build a workflow that automatically sorts incoming emails via AI into one of four categories, Urgent, Follow-up, Info, or Spam, applies the matching label in the email tool, sends an immediate notification when something is Urgent, and optionally dispatches a daily short report.

Adapt the structure to the concepts of your automation tool, for example nodes, steps, zaps, or agent actions. While doing so, preserve the trigger logic, the sequence, the parallelism, and the data flow.

TRIGGER
Type: a new incoming email, for example in Gmail or Outlook. Alternatively, scheduled hourly as a batch over all new emails since the last run.
Filter: only the desired inbox or label, for example the main inbox. Emails already categorized are skipped.

STEPS
1. Extract email data: sender, subject, content, and date received.
2. AI step: classification. Use the classification prompt below exactly. The output is JSON with the fields category, rationale, and priority. After this, step 3 and step 4 run in parallel.
3. Set label: the JSON field category is mapped to the label of the same name in the email tool, Urgent, Follow-up, Info, or Spam. The four labels must already exist in the tool.
4. Condition: only when category equals Urgent, a notification is triggered in Slack, Teams, or by email, with sender, subject, and rationale.
5. Output, optional: a daily inbox report. Use the summary prompt below exactly, over all emails categorized that day. Delivery by email or Slack.

PROMPTS, COPY EXACTLY AND UNCHANGED
Classification prompt:
"""
You are an email triage assistant. Address the user in a professional manner throughout. Analyze the following email and assign it to exactly one of the following categories. If information relevant to the classification is missing, mark your assessment as an assumption rather than presenting it as certain.

Email details:
Sender: [SENDER]
Subject: [SUBJECT]
Content: [EMAIL_CONTENT]
Received on: [DATE_RECEIVED]

Categories:
1. Urgent: time-critical requests, escalations, important customer communication, urgent appointments
2. Follow-up: emails that require a response or action but are not time-critical
3. Info: newsletters, notifications, FYI emails with no action required
4. Spam: advertising, unwanted emails, newsletters from unknown senders

Context of your work:
[WORK_CONTEXT]

Output format, JSON only:
{
  "category": "Urgent|Follow-up|Info|Spam",
  "rationale": "Short rationale in 1 sentence",
  "priority": "high|medium|low"
}
"""

Summary prompt:
"""
Create a concise summary of the email triage results, addressing the user in a professional manner throughout.

Categorized emails:
[EMAIL_LIST]

Statistics:
Total: [TOTAL_COUNT]
Urgent: [URGENT_COUNT]
Follow-up: [FOLLOWUP_COUNT]
Info: [INFO_COUNT]
Spam: [SPAM_COUNT]

Output format:
Create a structured summary with:
1. Heading "Email triage [DATE/TIME]"
2. Statistics overview
3. The three most urgent emails with sender and subject
4. A recommendation on which emails to handle first

Keep the summary to a maximum of 200 words.
"""

DATA FLOW
Step 1 to step 2: [SENDER], [SUBJECT], [EMAIL_CONTENT], [DATE_RECEIVED].
Step 2 to step 3 and 4: the JSON fields category, rationale, and priority.
Daily run to step 5: [EMAIL_LIST], [TOTAL_COUNT], [URGENT_COUNT], [FOLLOWUP_COUNT], [INFO_COUNT], [SPAM_COUNT].

ERROR HANDLING
AI returns no valid JSON: retry once, then set the default category Follow-up and log the error.
Label does not exist: log the error, do not abort the workflow. Create the labels in advance.
Notification fails: check the webhook URL or channel ID, log the error.
No write access in the email tool: report it as an open configuration item.

PLACEHOLDER CONVENTION
All placeholders in square brackets are variables. Map them dynamically to data from previous steps, as described in the data flow. Exception: [WORK_CONTEXT] is not dynamic. Set it manually before activation, with role, topics, and important senders. List such manual placeholders separately at the end.

SELF CHECK BEFORE COMPLETION
Check, before you report the workflow as finished:
1. Are all steps connected correctly, including the parallelism of step 3 and 4 and the Urgent condition?
2. Are both prompts inserted completely and unchanged, and are all placeholders mapped?
3. Is the JSON output from step 2 parsed correctly, with category, rationale, and priority?
4. Are all points listed that require manual configuration: inbox connection, LLM API key, channel ID, the four labels, and [WORK_CONTEXT]?

Working prompts

You are an email triage assistant. Address the user in a professional manner throughout. Analyze the following email and assign it to exactly one of the following categories. If information relevant to the classification is missing, mark your assessment as an assumption rather than presenting it as certain.

Email details:
Sender: [SENDER]
Subject: [SUBJECT]
Content: [EMAIL_CONTENT]
Received on: [DATE_RECEIVED]

Categories:
1. Urgent: time-critical requests, escalations, important customer communication, urgent appointments
2. Follow-up: emails that require a response or action but are not time-critical
3. Info: newsletters, notifications, FYI emails with no action required
4. Spam: advertising, unwanted emails, newsletters from unknown senders

Context of your work:
[WORK_CONTEXT]

Output format, JSON only:
{
  "category": "Urgent|Follow-up|Info|Spam",
  "rationale": "Short rationale in 1 sentence",
  "priority": "high|medium|low"
}

Create a concise summary of the email triage results, addressing the user in a professional manner throughout.

Categorized emails:
[EMAIL_LIST]

Statistics:
Total: [TOTAL_COUNT]
Urgent: [URGENT_COUNT]
Follow-up: [FOLLOWUP_COUNT]
Info: [INFO_COUNT]
Spam: [SPAM_COUNT]

Output format:
Create a structured summary with:
1. Heading "Email triage [DATE/TIME]"
2. Statistics overview
3. The three most urgent emails with sender and subject
4. A recommendation on which emails to handle first

Keep the summary to a maximum of 200 words.

JSON export