All posts
4 min read
RapidClaw Team The team behind RapidClaw.app

How I Automated My Email Triage with a Telegram AI Agent

I built an AI agent that scans my Gmail every hour, categorizes everything, and sends me the 3 most urgent items in Telegram. Here's how.

How I Automated My Email Triage with a Telegram AI Agent

I used to start every morning staring at 80+ unread emails. Client requests buried under newsletters, receipts, and LinkedIn notifications. I'd spend 45 minutes just figuring out what actually mattered. Then I'd do it again after lunch. And again at 5pm.

One Tuesday, I missed a time-sensitive contract renewal because it landed between a Figma update email and a SaaS receipt. That was the last straw.

How do you deal with email overload without an assistant?#

Most people do some version of this: open Gmail, scan subjects, mentally sort into "deal with now" / "deal with later" / "ignore." Maybe you've set up filters. Maybe you use labels. Maybe you've tried Superhuman or SaneBox or one of the dozen inbox tools.

The problem isn't any single email. It's the volume. A consultant juggling 3-4 clients gets 100+ emails a day. About 10 of those actually matter. The rest are noise. But you can't just ignore your inbox because the 10 important ones are mixed in randomly.

Filters help but they're static. A client who normally sends FYI updates might suddenly send something urgent. A new contact won't match any filter. And labels don't tell you what to do first.

What I actually needed was someone to read everything, sort it by urgency, and tap me on the shoulder when something needed my attention. Like a human assistant, minus the $4K/month.

What does an email triage agent actually do?#

Here's what my agent does every hour between 8am and 8pm:

1. Scans my Gmail inbox using the Gmail skill (OAuth connection, read-only access). It pulls every unread message that arrived since the last scan.

2. Categorizes each email into four buckets:

  • Urgent — needs a response within hours. Client escalations, time-sensitive requests, anything with a deadline today.
  • Respond Today — important but not on fire. Meeting confirmations, project questions, proposals to review.
  • FYI — good to know, no action needed. Team updates, shared docs, informational CCs.
  • Trash — newsletters, marketing, automated notifications I never read.

3. Sends me a Telegram message with the top 3 urgent items. Each one gets a one-line summary so I can decide in seconds whether to open the email or not.

Here's what a typical hourly update looks like:

🔴 URGENT (2)
1. Sarah (Acme Corp) — Asking for revised SOW by EOD tomorrow
2. AWS — Billing alert: $847 charge on production account

🟡 RESPOND TODAY (3)
1. Marcus — Wants to schedule Q2 planning call
2. Legal — Contract draft ready for review
3. Jen (Startup Y) — Feedback on prototype, non-blocking

📬 12 FYI archived | 8 trash archived

4. Drafts responses for "respond today" items. I get the draft in Telegram, edit it if needed, and tell the agent to send it. I never open Gmail for routine replies anymore.

5. End-of-day recap at 8pm:

📊 Email Day Summary
- 5 emails responded to
- 12 archived as FYI
- 8 auto-trashed
- 3 flagged for tomorrow
- Avg response time: 47 minutes (down from 4+ hours)

The agent also learns over time. After two weeks, it figured out that emails from my accountant are always urgent during tax season but FYI the rest of the year. It knows that GitHub notification emails are always FYI unless they mention a production incident.

This pattern recognition is the part that static filters can never do. The agent understands context, not just keywords.

How do you set this up?#

On RapidClaw, you'd pick a template or start from scratch. Connect your Gmail (OAuth — the agent gets read access, never deletes anything without your say-so). Set your scan schedule. Activate.

The whole thing takes about 2 minutes. You get a Telegram bot that starts scanning on the next scheduled interval.

No server to manage. No API keys to juggle (AI credits are included in your plan). No Docker containers to babysit. The agent runs 24/7 on RapidClaw's infrastructure.

If you want to customize the categories or add rules like "anything from @acme.com is always urgent," you just tell the agent in plain English through Telegram. It updates its own behavior.

Who is this for?#

This is built for anyone who gets more than 50 emails a day and can't afford to miss the important ones.

Consultants managing multiple clients. Founders who are the bottleneck on approvals. Agency owners who need to stay responsive but can't live in their inbox. Project managers getting CC'd on everything.

If you've ever thought about hiring a virtual assistant just to sort your email, this does that job for a fraction of the cost. And it doesn't take PTO.

How much does this cost?#

A human VA for email management runs $1,500-4,000/month. SaneBox is $7/month but only does filtering — no summaries, no drafts, no Telegram delivery. Superhuman is $30/month but you still have to open the app and do the work yourself.

RapidClaw starts at $19/month with AI credits included. That gets you the email triage agent plus any other agents you want to run. The Pro plan at $39/month gives you more AI credits for heavier scanning volumes.

Frequently Asked Questions#

How long does it take to set up an email triage agent?#

Under 2 minutes. Connect your Gmail account via OAuth, set your scan schedule (I do hourly from 8am-8pm), and activate. The agent starts categorizing on the next interval. No code, no configuration files.

Can the agent accidentally delete important emails?#

No. The agent has read-only access by default. It never deletes, moves, or modifies your emails without explicit approval. When it "archives" something, it's marking it in its own memory — your Gmail inbox stays untouched unless you tell it otherwise.

Does it work with Outlook or other email providers?#

Right now the primary integration is Gmail via OAuth. Outlook/Microsoft 365 support via the Microsoft Graph API is on the roadmap. If you're an Outlook user, you can still use the agent — you'd forward emails to a monitored address, though the native experience is smoother with Gmail.

How accurate is the categorization?#

In my experience, it's about 90% accurate from day one and improves to 95%+ within two weeks as it learns your patterns. You can correct it by saying things like "emails from @acme.com should always be urgent" and it adjusts immediately.

Can I use this with Discord instead of Telegram?#

Yes. RapidClaw agents work on both Telegram and Discord. The email summaries, drafts, and recaps work identically on either platform. Most people pick whichever they already have open all day.

Share this post

Ready to build your own AI agent?

Deploy a personal AI agent to Telegram or Discord in 60 seconds. From $19/mo.

Get Started

Stay in the loop

New use cases, product updates, and guides. No spam.