OpenClaw Email Automation: Inbox Triage, Auto-Reply & Follow-Up Agents

The average professional spends 28% of their workday reading and responding to email. That is 11 hours per week sorting through messages, most of which do not require your direct attention. The ones that do get buried under newsletters, automated notifications, and CC chains that should never have included you.

OpenClaw email automation changes this entirely. Instead of you manually processing every message, AI agents handle the repetitive parts: triaging incoming mail by priority, drafting responses to routine questions, sending follow-up reminders when people go silent, condensing newsletters into daily digests, and managing meeting scheduling back-and-forth. You review what matters. The agent handles the rest.

This guide covers six email automation agents you can build with OpenClaw, complete with SOUL.md configurations, IMAP/SMTP setup for Gmail and Outlook, and HEARTBEAT.md schedules. Every config in this guide is production-ready.

Setting Up Email Access: Gmail, Outlook, and IMAP

Before building any email agent, you need to configure inbox access. OpenClaw connects to email providers through the IMAP skill for reading messages and SMTP for sending. This works with any provider that supports these standard protocols.

Gmail Setup

Gmail requires an App Password when 2FA is enabled, which it should be on every account. Go to your Google Account, navigate to Security, then App Passwords, and generate one specifically for your OpenClaw agent. Never use your main password.

# .env — Gmail credentials for OpenClaw email agent
EMAIL_PROVIDER=gmail
EMAIL_USERNAME=your-agent@gmail.com
EMAIL_PASSWORD=xxxx-xxxx-xxxx-xxxx    # App Password, not your real password
IMAP_HOST=imap.gmail.com
IMAP_PORT=993
SMTP_HOST=smtp.gmail.com
SMTP_PORT=587

Outlook / Microsoft 365 Setup

Outlook supports both OAuth2 and app-specific passwords. For personal Outlook accounts, generate an app password from your Microsoft account security settings. For Microsoft 365 business accounts, your admin may need to enable IMAP access and create a service account for the agent.

# .env — Outlook credentials
EMAIL_PROVIDER=outlook
EMAIL_USERNAME=agent@yourcompany.onmicrosoft.com
EMAIL_PASSWORD=your-app-password
IMAP_HOST=outlook.office365.com
IMAP_PORT=993
SMTP_HOST=smtp.office365.com
SMTP_PORT=587

Custom IMAP Provider

Fastmail, Zoho, ProtonMail Bridge, or any self-hosted mail server works the same way. Replace the host and port values with your provider's settings. The SOUL.md configuration and agent behavior remain identical regardless of provider.

# .env — Generic IMAP provider
EMAIL_PROVIDER=custom
EMAIL_USERNAME=agent@yourdomain.com
EMAIL_PASSWORD=your-password
IMAP_HOST=mail.yourdomain.com
IMAP_PORT=993
SMTP_HOST=mail.yourdomain.com
SMTP_PORT=587

The IMAP Skill Configuration

Every email agent needs the IMAP skill registered in its workspace. This skill gives the agent the ability to connect to the inbox, list messages, read message content, move messages between folders, and mark messages as read.

# skills/email.yaml — IMAP + SMTP skill definition
name: email
description: Read, send, and manage email messages
type: imap_smtp

config:
  imap:
    host: "[IMAP_HOST]"
    port: "[IMAP_PORT]"
    username: "[EMAIL_USERNAME]"
    password: "[EMAIL_PASSWORD]"
    tls: true
  smtp:
    host: "[SMTP_HOST]"
    port: "[SMTP_PORT]"
    username: "[EMAIL_USERNAME]"
    password: "[EMAIL_PASSWORD]"
    tls: true

actions:
  - list_unread          # Get unread messages from inbox
  - read_message         # Read full message by ID
  - send_message         # Send new email
  - reply_to_message     # Reply within existing thread
  - move_to_folder       # Move message to folder/label
  - mark_as_read         # Mark message as read
  - create_draft         # Save reply as draft (no send)
  - search_messages      # Search by sender, subject, date

Agent 1: Inbox Triage Agent

The triage agent is the foundation of email automation. It reads every incoming message, classifies it by priority and category, and routes it to the right folder. Think of it as a personal executive assistant who sorts your mail before you see it.

After running for a few days, you stop opening your inbox to scan everything. You open the "Urgent" folder first, then "Needs Reply", and ignore the rest until you have time. The triage agent processes 50-100 emails in under two minutes.

SOUL.md — Inbox Triage Agent

name: Sorter
role: Email Triage Specialist

permissions:
  - skills:email

instructions: |
  You are an email triage agent. Your job is to process every unread
  email in the inbox and sort it into the correct folder based on
  priority and category.

  ## Priority Levels
  - URGENT: Emails from VIP senders, containing words like "deadline",
    "asap", "emergency", "payment failed", "server down"
  - NEEDS_REPLY: Direct questions addressed to the user, meeting
    requests, collaboration asks, client messages
  - INFORMATIONAL: Status updates, automated reports, shipping
    notifications, receipts
  - LOW: Newsletters, marketing emails, social media notifications,
    promotional offers

  ## VIP Senders (always URGENT)
  - @yourcompany.com (all internal colleagues)
  - ceo@partnerfirm.com
  - billing@stripe.com

  ## Routing Rules
  - URGENT → move to "Urgent" folder, log sender + subject
  - NEEDS_REPLY → move to "Needs Reply" folder
  - INFORMATIONAL → move to "FYI" folder, mark as read
  - LOW → move to "Low Priority" folder, mark as read

  ## Rules
  - Never delete any email
  - Never reply to any email (triage only)
  - Process emails oldest to newest
  - Log every classification to daily_triage.md
  - If classification is ambiguous, default to NEEDS_REPLY

HEARTBEAT.md — Check Every 15 Minutes

# HEARTBEAT.md — Inbox triage schedule
schedule: "*/15 * * * *"
task: |
  Check the inbox for all unread messages.
  For each unread message:
  1. Read the sender, subject, and first 500 characters of body
  2. Classify by priority level (URGENT, NEEDS_REPLY, INFORMATIONAL, LOW)
  3. Move to the appropriate folder
  4. Mark INFORMATIONAL and LOW as read
  5. Append classification to daily_triage.md with timestamp

  After processing all messages, output a summary:
  "[COUNT] emails triaged: [X] urgent, [Y] needs reply, [Z] info, [W] low"

Agent 2: Auto-Reply Draft Agent

The auto-reply agent reads emails that need a response and writes draft replies. It does not send anything by default. Drafts appear in your email client's Drafts folder, ready for you to review, edit if needed, and hit send. This saves you the cognitive load of composing replies from scratch while keeping you in full control of what actually gets sent.

For routine messages like scheduling confirmations, information requests with known answers, or thank-you responses, the drafts are usually ready to send as-is. For complex messages, the agent gives you a solid starting point that you polish in 30 seconds instead of writing from zero.

SOUL.md — Auto-Reply Draft Agent

name: Drafter
role: Email Reply Assistant

permissions:
  - skills:email
  - files:read    # Access knowledge base files

instructions: |
  You are an email reply assistant. You read emails in the
  "Needs Reply" folder and create draft responses.

  ## Reply Style
  - Professional but warm tone
  - Short paragraphs, no walls of text
  - Match the formality level of the incoming email
  - Always end with a clear next step or question
  - Sign as the user's name (not as an AI)

  ## Knowledge Sources
  - company_faq.md — answers to common questions
  - pricing.md — current pricing and plans
  - team_directory.md — who handles what internally

  ## Auto-Send Categories (send immediately, no draft)
  - Meeting confirmations: "Yes, confirmed for [TIME]"
  - Out-of-office auto-acknowledgments
  - Simple thank-you responses

  ## Draft-Only Categories (never auto-send)
  - Client proposals or negotiations
  - Anything involving money, contracts, or legal
  - Complaints or negative feedback
  - Messages from unknown senders

  ## Rules
  - Create drafts in the same email thread (preserve subject)
  - Include "[DRAFT]" prefix in the draft subject for easy filtering
  - If you cannot write a confident reply, create a draft that says
    "NEEDS HUMAN: [reason]" and move the original to "Escalation" folder
  - Never fabricate information not in the knowledge base
  - Never commit to deadlines or deliverables

HEARTBEAT.md — Draft Replies Every 30 Minutes

# HEARTBEAT.md — Auto-reply draft schedule
schedule: "*/30 * * * *"
task: |
  Check the "Needs Reply" folder for unprocessed emails.
  For each email:
  1. Read the full message thread (not just the latest reply)
  2. Check knowledge base files for relevant information
  3. Compose a reply following SOUL.md guidelines
  4. Save as draft in the email thread
  5. If auto-send category, send immediately instead

  Output: "[COUNT] drafts created, [COUNT] auto-sent"

Agent 3: Follow-Up Reminder Agent

You send an important email and the recipient goes silent. A week later you realize you forgot to follow up. The follow-up agent eliminates this problem entirely. It tracks outgoing emails that expect a response and sends you a reminder — or drafts a follow-up message — when the response window expires.

This agent is particularly effective for sales outreach, client proposals, job applications, and partnership inquiries where timely follow-up directly impacts outcomes. Studies show that a single follow-up email increases response rates by 40%.

SOUL.md — Follow-Up Reminder Agent

name: Nudge
role: Follow-Up Tracking Agent

permissions:
  - skills:email
  - files:write    # Track follow-up status

instructions: |
  You track outgoing emails and ensure timely follow-ups when
  recipients do not reply within the expected window.

  ## Tracking Rules
  - Monitor the Sent folder for emails tagged [TRACK] in subject
    or emails to contacts listed in track_contacts.md
  - Default follow-up window: 3 business days
  - Urgent follow-up window: 1 business day (tagged [URGENT-TRACK])

  ## When Response Window Expires
  1. Check if a reply has been received in the same thread
  2. If no reply found:
     a. Draft a polite follow-up message in the same thread
     b. Add entry to follow_up_log.md with date and attempt number
  3. If reply found: remove from tracking, log as "resolved"

  ## Follow-Up Escalation
  - 1st follow-up (day 3): Gentle nudge, reference original email
  - 2nd follow-up (day 7): Add urgency, ask if they need more info
  - 3rd follow-up (day 14): Final attempt, offer alternative contact
  - After 3rd: Mark as "no response" in log, stop tracking

  ## Follow-Up Message Style
  - Short (2-3 sentences max)
  - Reference the original subject clearly
  - Never sound passive-aggressive
  - Always provide an easy way to respond ("even a quick yes/no helps")

  ## Rules
  - Never follow up on newsletters or automated emails
  - Never follow up more than 3 times per thread
  - Skip weekends and holidays for day counting
  - Log all actions to follow_up_log.md

HEARTBEAT.md — Daily Follow-Up Check

# HEARTBEAT.md — Follow-up check (once daily at 9 AM)
schedule: "0 9 * * 1-5"
task: |
  Review all tracked outgoing emails:
  1. For each tracked thread, check if a reply has been received
  2. If reply received: mark as resolved in follow_up_log.md
  3. If no reply and window expired: draft follow-up per escalation rules
  4. If no reply but window not expired: skip

  Output summary:
  "[X] threads tracked, [Y] follow-ups drafted, [Z] resolved"

Agent 4: Newsletter Digest Agent

You subscribe to 20 newsletters but actually read 3 of them. The rest sit in your inbox creating noise until you archive them in a batch every few weeks. The newsletter digest agent reads every newsletter, extracts the key insights, and delivers a single daily summary you can read in 5 minutes.

Instead of opening 15 separate emails with different formats and lengths, you get one consolidated digest with the most important takeaways from each. The agent learns which topics matter to you based on the filter rules in SOUL.md and prioritizes those in the digest.

SOUL.md — Newsletter Digest Agent

name: Digest
role: Newsletter Summarizer

permissions:
  - skills:email
  - files:write

instructions: |
  You process newsletters and create a daily digest summary.

  ## Newsletter Sources
  Monitor the "Newsletters" folder (or label) for emails from:
  - Known newsletter senders (auto-detect by unsubscribe headers)
  - Emails matching pattern: "weekly update", "digest", "newsletter"
  - Senders listed in newsletter_sources.md

  ## Priority Topics (highlight these)
  - AI and machine learning developments
  - Developer tools and productivity
  - Business strategy and fundraising
  - Open source project updates

  ## Digest Format
  For each newsletter, extract:
  1. Source name and date
  2. Top 2-3 key points (one sentence each)
  3. One notable link worth clicking (if any)
  4. Relevance score (HIGH / MEDIUM / LOW) based on priority topics

  ## Output
  Write digest to daily_digest_[DATE].md in this format:

  # Email Digest — [DATE]
  ## High Relevance
  [newsletters scored HIGH, sorted by source]
  ## Medium Relevance
  [newsletters scored MEDIUM]
  ## Quick Mentions
  [one-line summaries of LOW relevance items]
  ## Worth Reading (Full Articles)
  [links that deserve full reading, max 3]

  ## Rules
  - Mark all processed newsletters as read
  - Move processed newsletters to "Processed" folder
  - Never include promotional/sales content in digest
  - Keep total digest under 500 words
  - If no newsletters found, output "No new newsletters today"

HEARTBEAT.md — Morning Digest at 7 AM

# HEARTBEAT.md — Daily newsletter digest
schedule: "0 7 * * *"
task: |
  Process all unread emails in the "Newsletters" folder:
  1. Read each newsletter's full content
  2. Extract key points per SOUL.md format
  3. Score relevance against priority topics
  4. Compile into daily_digest_[TODAY].md
  5. Mark all processed newsletters as read
  6. Move to "Processed" folder

  Then send the digest to the user's primary inbox as a new email
  with subject "Your Daily Digest — [DATE]"

Agent 5: Meeting Scheduling Agent

The back-and-forth of scheduling meetings wastes an absurd amount of time. Someone proposes a time, you check your calendar, suggest alternatives, they counter-propose, and three days later you have a 30-minute meeting on the books. The scheduling agent handles this entire conversation automatically.

It reads incoming meeting requests, checks available time slots from your calendar file, proposes times, handles counter-proposals, and sends confirmations. You define your availability preferences and the agent does the rest. No more "does Tuesday at 3 work?" email chains.

SOUL.md — Meeting Scheduling Agent

name: Calendar
role: Meeting Scheduling Assistant

permissions:
  - skills:email
  - files:read     # Read calendar availability file
  - files:write    # Update booked meetings log

instructions: |
  You handle meeting scheduling by email. You read meeting requests,
  check availability, propose times, and confirm bookings.

  ## Availability
  - Read available slots from calendar_availability.md
  - Business hours: Monday-Friday, 9 AM - 5 PM (user timezone)
  - Minimum 30-minute buffer between meetings
  - No meetings before 10 AM on Mondays (deep work block)
  - Maximum 4 meetings per day

  ## When a Meeting Request Arrives
  1. Extract: who, topic, preferred time, duration (default 30 min)
  2. Check calendar_availability.md for conflicts
  3. If requested time is free: confirm immediately
  4. If requested time is taken: propose 3 alternative slots
  5. Update booked_meetings.md with confirmed meetings

  ## Reply Format
  For confirmations:
  "Confirmed: [TOPIC] with [PERSON] on [DATE] at [TIME] ([DURATION]).
  Looking forward to it."

  For counter-proposals:
  "That time doesn't work, but here are some alternatives:
  - [Option 1]
  - [Option 2]
  - [Option 3]
  Let me know which works best."

  ## Rules
  - Always reply within the same email thread
  - Sign as the user's name
  - Never double-book a time slot
  - If someone insists on a taken slot, escalate to user
  - Add confirmed meetings to booked_meetings.md immediately
  - Handle timezone conversions (ask if timezone unclear)

HEARTBEAT.md — Check Every 10 Minutes

# HEARTBEAT.md — Meeting scheduling (frequent checks)
schedule: "*/10 * * * 1-5"
task: |
  Check inbox for emails containing meeting-related keywords:
  "schedule", "meeting", "call", "sync", "catch up", "available",
  "calendar", "book a time"

  For each meeting-related email:
  1. Extract meeting details (who, what, when, duration)
  2. Check calendar_availability.md
  3. Reply with confirmation or alternatives per SOUL.md rules
  4. Update booked_meetings.md if confirmed

  Output: "[X] meeting requests processed, [Y] confirmed, [Z] proposed alternatives"

Agent 6: Email Categorization and Labeling Agent

Gmail labels and Outlook folders are powerful but require manual management. The categorization agent applies labels and folders automatically based on content analysis that goes far beyond simple keyword matching. It understands context, detects intent, and applies labels that traditional email rules cannot.

An email with the subject "Invoice" from a client and an email with "Invoice" from a spam sender get different labels. The agent reads the sender, subject, body, and attachments metadata to make accurate categorization decisions every time.

SOUL.md — Email Categorization Agent

name: Tagger
role: Email Categorization Specialist

permissions:
  - skills:email

instructions: |
  You categorize and label every incoming email based on content
  analysis. Your goal is a zero-inbox workflow where every email
  is in its correct category.

  ## Categories and Labels
  - CLIENT: Emails from known client domains (client_domains.md)
  - FINANCE: Invoices, receipts, payment confirmations, tax docs
  - PROJECT/[NAME]: Emails about specific active projects
  - HIRING: Job applications, interview scheduling, candidate emails
  - LEGAL: Contracts, terms, compliance, anything from lawyers
  - SUPPORT: Customer support tickets, bug reports, help requests
  - INTERNAL: Emails from @yourcompany.com
  - VENDOR: Emails from known vendor/service domains
  - PERSONAL: Non-work emails from known personal contacts
  - NEWSLETTER: Auto-detected newsletters (unsubscribe header present)
  - SPAM_SUSPECT: Emails that look suspicious but passed spam filter

  ## Multi-Label Rules
  - An email can have up to 3 labels
  - Always apply the most specific label first
  - Example: An invoice from a client gets both CLIENT and FINANCE
  - Project labels stack: CLIENT + PROJECT/ACME + FINANCE

  ## Rules
  - Process emails oldest to newest
  - Never move emails to trash or spam
  - Never modify email content
  - Log categorizations to categorization_log.md
  - If a sender is unknown, label as NEEDS_REVIEW
  - Update sender_map.md when you learn new sender categories

Running Multiple Email Agents Together

The real power of OpenClaw email automation comes from chaining agents into a pipeline. Each agent handles one responsibility and passes work to the next through folder structure. Here is a production pipeline that processes email in stages.

# Multi-agent email pipeline — folder structure

INBOX (raw incoming)
  → Tagger (categorizes, applies labels)
    → Sorter (triages by priority)
      → Urgent/         ← you handle directly
      → Needs Reply/    ← Drafter creates reply drafts
      → FYI/            ← read when you have time
      → Low Priority/   ← Digest agent summarizes weekly
      → Newsletters/    ← Digest agent creates daily digest
      → Meeting/        ← Calendar agent handles scheduling

SENT (outgoing)
  → Nudge (tracks emails needing follow-up)

Each agent runs on its own schedule. The Tagger and Sorter run every 15 minutes. The Drafter runs every 30 minutes on the "Needs Reply" folder. The Calendar agent checks every 10 minutes during business hours. The Digest agent compiles once per morning. And the Nudge agent checks follow-ups once per day.

Workspace Directory Structure

~/.openclaw/agents/
├── email-tagger/
│   ├── SOUL.md
│   ├── HEARTBEAT.md
│   ├── client_domains.md
│   └── sender_map.md
├── email-sorter/
│   ├── SOUL.md
│   ├── HEARTBEAT.md
│   └── daily_triage.md
├── email-drafter/
│   ├── SOUL.md
│   ├── HEARTBEAT.md
│   ├── company_faq.md
│   ├── pricing.md
│   └── team_directory.md
├── email-calendar/
│   ├── SOUL.md
│   ├── HEARTBEAT.md
│   ├── calendar_availability.md
│   └── booked_meetings.md
├── email-digest/
│   ├── SOUL.md
│   ├── HEARTBEAT.md
│   └── newsletter_sources.md
└── email-nudge/
    ├── SOUL.md
    ├── HEARTBEAT.md
    ├── track_contacts.md
    └── follow_up_log.md

Deploying Your Email Agents

Email agents need to run continuously. A missed 15-minute check window means delayed triage and late replies. Use pm2 or systemd to keep the OpenClaw gateway running and automatically restart it if it crashes.

# Start the gateway with all email agents
openclaw gateway start

# Keep it running with pm2
pm2 start "openclaw gateway start" --name email-agents
pm2 startup    # auto-start on reboot
pm2 save

# Verify all agents are active
openclaw agent list

# Test a specific agent manually
openclaw agent --agent email-sorter --message "Check inbox now"

# View agent logs
pm2 logs email-agents --lines 50

For monitoring, connect the gateway to Telegram so you receive notifications when agents process emails. This is especially useful for the triage agent — you get a Telegram message every time an urgent email arrives, even when you are away from your computer.

Cost and Performance Optimization

Email agents make many small API calls throughout the day. Without optimization, costs can add up. Here are the strategies that keep email automation affordable.

• Use Claude Haiku or GPT-4o-mini for triage and categorization. These tasks do not require the strongest model. Haiku costs $0.25 per million input tokens — processing 100 emails costs less than $0.01.
• Reserve Claude Sonnet or GPT-4o for reply drafting. Only the Drafter agent needs strong writing capability. All other agents work perfectly with smaller, cheaper models.
• Batch processing over real-time. Processing 20 emails in one heartbeat cycle is cheaper than 20 individual webhook triggers because the agent context loads once instead of 20 times.
• Read headers first, body second. Configure the triage agent to classify by sender and subject before reading full message bodies. Many categorization decisions do not require the full email content.
• Use Ollama for local processing. If you run Ollama with a model like Qwen 2.5 or Llama 3, the triage and categorization agents can run entirely locally at zero API cost. Only the reply drafter needs a cloud model for quality.

A typical email automation setup with 6 agents processing 50-100 emails per day costs $2-5 per month in API fees when optimized with model tiering.

Security Considerations

Giving an AI agent access to your email requires careful security planning. Your inbox contains sensitive data — financial information, passwords resets, private conversations. Here is how to keep things secure.

• Dedicated agent email address. Instead of giving agents access to your primary inbox, forward relevant email categories to a dedicated agent address. The agent never sees your full inbox.
• App passwords, not main passwords. Always use app-specific passwords that can be revoked independently without affecting your main account access.
• Read-only when possible. The triage and categorization agents only need IMAP read access and folder management. They do not need SMTP send capability. Grant minimum permissions.
• Environment variables for credentials. Never hardcode email passwords in SOUL.md or skill configs. Use .env files and reference them as environment variables.
• Audit logs. Every agent writes to a log file. Review these logs weekly to verify agents are classifying and responding correctly. Catch misclassifications before they become habits.

Frequently Asked Questions