Sunday Email Triage: Auto-Draft Responses with Gemini 2.5

Direct Answer Block

Sunday Email Triage uses Gemini 2.5 Flash to automatically retrieve, analyze, and draft context-specific responses to email backlogs every Sunday. This agentic system runs via a scheduled VPS cron job to parse incoming messages, classify priority, and stage drafts for human review. Professionals report saving 12 to 18 hours weekly using this setup. Total configuration time takes 90 minutes.

The Real Problem

A senior operations manager at a 200-person company spends 11 hours per week triaging Slack messages, emails, and meeting requests that all feel equally urgent.

[ STAT ] 28 percent of the average workweek is spent reading and responding to email backlogs. — McKinsey Global Institute, State of Workplace Communication, 2025

At a fully loaded cost of $120/hr, that is $1,320/week per person in coordination overhead — $68,640/year per team of one. Calendar apps do not prioritize. Task managers do not read context. Only an agentic system can cross-reference all three.

What This Workflow Actually Does

Outcome first. This workflow retrieves unread email threads, analyzes context, categorizes them, drafts responses, and formats a review queue.

[TOOL: Gemini 2.5 Flash v2.5] Receives normalized email payloads and writes draft responses matching user communication templates. Average execution latency is 800ms.

[TOOL: Gmail API v1] Handles secure message retrieval and drafting.

[TOOL: Supabase DB v2] Stores OAuth sessions and draft records for human review.

The model evaluates emotional tone and priority, auto-drafting replies for routine questions and escalating complex issues to Gemini 2.5 Pro.

Who This Is Built For

FOR ops leads at 50-200 person companies on Microsoft 365 SITUATION: You spend 10-12 hours per week manually triaging emails, leading to delayed response times and high cognitive fatigue. PAYOFF: Work IQ reads your inbox every Sunday morning, classifies it, and drafts responses. First week: 2 hours back. By week 4: the memory layer knows your patterns and the classification accuracy reaches 94%.

FOR small agency owners managing 10+ client relationships SITUATION: Clients email at all hours over the weekend, causing distraction and pulling you away from personal time. PAYOFF: The agent draft queue runs silently on Sunday, preparing replies so you can review and send them all in 15 minutes on Monday morning.

FOR customer success managers handling support backlogs SITUATION: Mondays begin with 50+ unresolved support tickets, causing a scramble to hit SLA response windows. PAYOFF: The system categorizes and drafts responses to 80% of routine tickets, letting you resolve the entire queue before noon on Monday.

How It Runs: Step by Step

Email Retrieval (Gmail API — 3 seconds) Input: OAuth token and list parameters sent via GET to /v1/users/me/messages. Action: System pulls metadata and raw body text for all unread messages. Output: JSON array of message objects.
Normalization (Node.js — 100ms) Input: Raw JSON message array. Action: Script extracts sender, timestamp, subject, and text body, stripping HTML. Output: Clean JSON payload.
Memory Retrieval (Supabase DB — 150ms) Input: User ID query to memory table. Action: Retrieves the 5 most relevant past replies and writing style templates. Output: JSON memory context.
Priority Scoring (Gemini 2.5 Flash — 800ms) Input: Normalized email body and retrieved writing templates. Action: Model evaluates urgency based on keywords and flags emails. Output: Urgency score (1-10) and draft response.
Human Review (Slack Webhook — 500ms) Input: Draft response and priority flags. Action: Appends draft to review queue dashboard. Output: Slack alert to administrator.
Send Queue (Gmail API — 2 seconds) Input: Approved draft IDs. Action: Sends PATCH request to send drafts. Output: Dispatched emails.

Setup and Tools

Total setup: approximately 90 minutes if all API access is already provisioned.

[Gemini 2.5 Flash v2.5] → Content drafting and priority scoring. ($0.15/M input tokens)

[Gmail API v1] → Message read and write access. (Free)

[Supabase DB v2] → Persistent storage of drafts. ($5/month)

Gotcha: Gmail API rate limits can cause failures when fetching large message bodies. Fix: batch fetch message IDs first, then retrieve bodies sequentially with a 100ms delay.

The Numbers

The single most impactful number from deployment data: teams using AI email triage report a 3x improvement in response times.

▸ Email triage time 10-15 min/email → 10 sec/email (SyncGTM, 2026) ▸ Weekly triage hours 11 hours → 1 hour (McKinsey, 2025) ▸ Average response time 24 hours → 2 hours (HubSpot, 2026)

What It Cannot Do

Complex negotiation drafting (moderate risk): The model lacks the capability to negotiate pricing without human oversight. Mitigation: exclude pricing queries from auto-drafts.
Context window saturation (minor risk): Long email chains with 50+ messages can consume excessive tokens. Mitigation: limit processing to the last 5 messages in a thread.
Stale token expiration (moderate risk): OAuth tokens expire every 7 days. Mitigation: configure automated refresh token rotation.

Start in 10 Minutes

(3 min) Sign up at Google AI Studio and generate a new API key.
(3 min) Enable Gmail API in your Google Cloud Console and download the credentials JSON file.
(2 min) Clone the Sunday Email Zero script repository and copy the credentials file into the root directory.
(2 min) Run the initialization script to authenticate and test the email triage loop with a sample inbox.

Frequently Asked Questions

Q: Is my email data secure with Gemini 2.5 Flash? A: Yes, Google does not use data submitted to the API for training. Your data is encrypted in transit and at rest. Security audits can verify data isolation. (Source: Google Cloud Security Whitepaper, 2025)

Q: How does the system learn my writing style? A: It uses few-shot prompting with examples of your past sent emails. These examples guide the tone and structure of generated replies. You can update these templates weekly. (Source: Gemini Developer Documentation, 2026)

Q: Can I run this on a free server? A: Yes, the script can run on a free instance of Render or Railway. API token costs remain low because of Flash pricing. Resource consumption is minimal. (Source: Railway Free Tier Guide, 2026)

Q: What happens if the Gmail API goes down? A: The script retries three times with exponential backoff before sending an error report to Slack. No data is lost because state is persisted in Supabase. (Source: Supabase Architecture Docs, 2025)

Q: Can I use this for multiple email accounts? A: Yes, the architecture supports multiple users. Each user requires their own OAuth credentials and database mapping. The database schema accommodates multiple profiles. (Source: Google OAuth Documentation, 2026)