Auto-Summarize Your Gmail Inbox Every Morning with n8n

Your inbox accumulates 40–80 emails overnight. By the time you've triaged them it's 9:30 AM and you haven't done a single thing that actually mattered. This workflow fixes that: at 7:00 AM every day, n8n pulls your unread Gmail, an AI node reads every subject and snippet, ranks what's genuinely urgent, and posts a 10-line briefing to wherever you're already looking — Slack, Telegram, whatever.

You'll build it once. It runs every day on its own. Here's exactly how.

---

What the Workflow Does

1. A Schedule Trigger fires at 07:00 every morning.
2. A Gmail node (OAuth2) fetches unread messages from the last 18 hours.
3. A Code node formats the emails into a structured prompt (sender, subject, first 200 characters of body).
4. An AI / LLM node (Basic LLM Chain) classifies and summarizes: "Needs reply today", "FYI only", "Can ignore".
5. A Set node shapes the output into a clean digest string.
6. A Slack or Telegram node posts the briefing.

Total nodes: 6. Build time: ~25 minutes.

---

Step-by-Step Build

1. Schedule Trigger

Drop in a Schedule Trigger node. Set:

• Trigger Interval: Days
• Hour: 7
• Minute: 0

This fires once per day. If you want weekdays only, add an IF node right after that checks {{ $now.weekday >= 1 && $now.weekday <= 5 }} and routes only the true branch forward. (In n8n, $now.weekday is a Luxon property — 1 is Monday, 7 is Sunday.)

2. Gmail Node — Fetch Unread Mail

Add a Gmail node set to Get Many messages. Config:

Operation:      Get Many
Filters:
  Read Status:  Unread
  Received After: {{ $now.minus({hours: 18}).toISO() }}
Limit:          50
Return All:     false

Connect your Gmail account via OAuth2 — n8n will prompt you to authorize once; after that it reuses the token silently.

Tip: Set the date filter to 18 hours (not 24) so you don't double-count emails from the previous summary if you run it slightly late one morning.

3. Code Node — Build the Prompt

An AI node needs the emails formatted as a single string. Use a Code node (JavaScript) to collapse the array:

const emails = $input.all();

const formatted = emails.map((item, i) => {
  const e = item.json;
  const snippet = (e.snippet || '').slice(0, 200);
  return `[${i + 1}] From: ${e.from} | Subject: ${e.subject}\n${snippet}`;
}).join('\n\n');

return [{ json: { emailBlock: formatted, count: emails.length } }];

This gives the AI a clean numbered list — much cheaper on tokens than sending raw MIME.

4. AI / LLM Node — Summarize and Rank

Add a Basic LLM Chain node. Connect it to the Code node's output.

In the Prompt field, use an expression like:

You are an executive assistant. Below are {{ $json.count }} unread emails from the last 18 hours.

Classify each as one of:
- URGENT: needs a reply or action today
- FYI: worth reading, no action
- NOISE: newsletters, notifications, can skip

Then write a 5–10 line morning briefing, starting with the URGENT items. Be concise.

Emails:
{{ $json.emailBlock }}

Leave Model at the default (the node will use whatever model is configured in your n8n credentials). On AgentRoost, those credentials already point to the included AI pool — no API key needed, no separate AI bill at the end of the month.

5. Set Node — Shape the Output

Add a Set node to extract just the AI's text for the next step:

Field:  digest
Value:  {{ $json.text }}

6. Deliver the Digest

Option A — Slack: Add a Slack node, action Send a Message. Set:

• Channel: #inbox-digest (or any private channel)
• Message: *Morning inbox briefing — {{ $now.toFormat('EEE d MMM') }}*\n\n{{ $json.digest }}

Option B — Telegram: Add a Telegram node, action Send Message. Use your bot credentials and paste the same message template. If you're on AgentRoost and also have a Hermes or OpenClaw agent, your Telegram bot is already provisioned — you can route to it directly.

---

Pitfalls and Tips

Gmail OAuth scopes. When authorizing, make sure gmail.readonly is included. n8n's Gmail OAuth2 credential requests it by default, but if you set up a custom Google Cloud app, verify it's listed under scopes.

Large inboxes. If you regularly get 100+ emails overnight, the prompt will get long. Add a Limit node or pre-filter in the Gmail node with label:important so only starred/important mail goes to the AI.

AI node "no output" errors. These usually mean the model returned an empty response because the prompt was too long. Add a Code node before the AI that truncates emailBlock to 4000 characters if emailBlock.length > 4000.

Credentials refresh. Gmail OAuth tokens expire. n8n handles the refresh automatically as long as the workflow runs at least once every 7 days. The Schedule Trigger ensures this — just don't pause the workflow for a week or your token goes stale.

Testing without waiting until 7 AM. Click Test Workflow in the editor. The Schedule Trigger will fire immediately in test mode; you'll see the live Gmail data and the AI summary right in the canvas.

---

How to Run This on AgentRoost

Here's what running this yourself on a VPS looks like: spin up a $6/mo server, install Node, install n8n, configure a process manager, set up SSL, sort out an OpenAI API key, watch the monthly AI bill, and remember to renew the server. The workflow itself takes 25 minutes — the plumbing takes a day.

On AgentRoost, you skip the plumbing entirely:

1. Sign up at agentroost.app — email, Google, Microsoft, or Discord.
2. Pick the n8n framework and name your instance (e.g., morning-digest).
3. Your private n8n editor opens at https://morning-digest.agentroost.app — it's yours. You own the login, the data, the workflows.
4. Build the 6-node workflow above. When you reach the AI node, the credentials are already wired to included AI credits. No API key dialog.
5. Activate the workflow. Done.

The instance runs 24/7 on dedicated hardware. The AI node works from day one because the LLM credits are included in the subscription — no separate OpenAI account, no bring-your-own-key. Plans start at $19.99/mo all-in, with a 14-day money-back guarantee and no annual lock-in.

Get started — pick the n8n plan →

---

Extending the Workflow

Once the basic digest is running, a few natural next steps:

• Reply drafts. Add a second AI node that outputs a draft reply for each URGENT email, then use the Gmail node's Reply action to save them as drafts.
• Calendar context. Add a Google Calendar node before the AI step. Inject today's meetings into the prompt so the AI can flag emails that conflict with or relate to scheduled calls.
• Filtering by sender. Add an IF node after the Gmail fetch to split VIP senders (your boss, key clients) into a priority lane — the AI sees those first and flags them explicitly.
• Weekly summary. Duplicate the workflow, change the Schedule Trigger to every Monday at 8 AM, and adjust the prompt to ask for a week-in-review instead of a morning briefing.