I built an email outreach engine into Hermes because we do a lot of link building at the agency and I wanted our own tool. Here’s exactly how the Hermes email skill works, the six steps behind it, and the safety rules I’d insist on before you point an agent at your inbox.
Short answer
- The Hermes email skill finds leads, enriches them, verifies emails, filters, sends and tracks replies.
- Six steps, each removing one reason cold email fails.
- You need Hunter, Google Workspace, Himalaya and Hermes agent.
- Use a separate inbox, keep draft mode on, and say it’s an AI agent.
What the Hermes email skill actually is
The Hermes email skill is an outreach engine that lives inside my agent operating system. It turns “I need to do some link building outreach” into a tab with a button, instead of a week of manual work in a spreadsheet.
I should be upfront: this is experimental. I built it because we do a lot of link building for clients at the agency and I wanted our own custom tool rather than paying for someone else’s forever. It works, it’s genuinely useful, and I’m still evolving it.
What it does end to end: you describe who you want to reach, it finds those companies, turns each one into a real named person with a real email, checks that email actually exists, throws out the companies that were never going to reply, sends from a dedicated inbox, and tracks who replied.
The six moving parts of the outreach engine
- Describe who you want. You type the target in plain English — “SEO agencies interested in white label link building” — and it goes and finds them.
- Enrich. This turns each company it finds into an actual person with contact details, rather than a generic info@ address nobody reads.
- Verify. It checks the email address is real before anything sends. This is the step that kills your bounce rate.
- Filter. It removes the giant companies that were never going to reply to a cold email anyway. Small and medium sites reply far more often.
- Send. One message becomes a drafted campaign, with a live preview, sent from a separate agent inbox.
- Track. Sent, drafts, bounces, contacted count and replies, all on one dashboard.
Each of those six steps removes one specific reason a cold email fails. That’s the whole design principle — it’s not clever, it’s just that most outreach dies from a bad address, a wrong-sized target, or a message that never got reviewed.
What the dashboard shows you
- Total leads, how many are validated, and how many are actually sendable.
- Companies flagged as too big to bother with, hidden from the list automatically.
- A recent activity log with the date, who was contacted and what the topic was.
- A lead table with domain, email, company size, status, whether it’s been enriched, and notes on each lead.
- Campaign stats — sent, drafts, bounces, and how many people you’ve contacted.
You can select which leads go into a campaign, and delete the ones you look at and think “no, we shouldn’t be emailing them”. Email addresses stay hidden on screen until you click to reveal them.
The four things you need to wire it together
| Component | What it does |
|---|---|
| Hermes agent | The hands — it runs the steps and does the work |
| Hunter API | Verifies email addresses actually exist before you send |
| Google Workspace API | Sends the mail and reads the inbox |
| Himalaya | A skill that lets the agent read the inbox and reply |
You can also plug in Firecrawl as a second API key for the lead-finding side. And you don’t need an expensive frontier model behind it — a Grok subscription or a MiniMax coding plan works fine, and so does OAuth login.
For how I built it: I used Claude Desktop. I gave it the idea, it helped me build the tool, then I plugged the finished thing into the agent operating system alongside all my other day-to-day workflows. If you want to see how that fits together, start with the Agentic OS dashboard.
The safety rules I’d insist on
This is an agent with access to email. Treat it that way.
- Give it a separate inbox. Not your main one. The agent shouldn’t have access to your real mail, and a dedicated inbox keeps the blast radius small if something goes wrong.
- Be clear it’s an AI agent. Put it in the email copy. People deserve to know, and it’s better for you than being found out.
- Use draft mode. Nothing sends until you approve it. You can review every single message first.
- Let the bounce protection do its job. If an address bounces too many times, the system stops rather than burning your domain reputation.
- Keep the stop button close. You can pause, stop or delete a campaign mid-flight if something reads wrong.
None of that is optional in my book. An outreach agent that sends unreviewed mail from your main inbox is a liability, not an asset.
Does cold email even still work?
I get this in the comments constantly. We build a lot of backlinks with cold email, so yes — but the version that works is targeted and verified, not blasted.
The other objection is that you need expensive tools. We do use Pitchbox at the agency, and it’s good. The point of building this was that with Hermes you can now have the same shape of system — find, verify, filter, send, track — running as your own tool, automated with AI.
The third objection is “I’ll set outreach up later when I have time.” Every week you wait is a week of no pipeline. Set up something small now and improve it.
Want the outreach engine? It’s one tab inside the Agent OS in the AI Profit Boardroom — full install file, every prompt, the memory setup, and coaching calls where we set up your outreach together. Or start free with the free AI course and community, and grab my link building book while you’re at it.
FAQ
What is the Hermes email skill?
An outreach engine built on Hermes that finds leads, enriches them into real contacts, verifies the email addresses, filters out oversized targets, sends from a dedicated inbox and tracks replies — all from one dashboard.
What APIs do I need?
Hunter for email verification, Google Workspace to send and read mail, Himalaya as the inbox-reading skill, and Hermes agent itself. Firecrawl is optional for lead finding.
Should I connect my main inbox?
No. Use a separate inbox for the agent. It keeps the agent away from your real mail and makes it obvious the outreach is coming from an automated system.
Will it send emails without me seeing them?
Not if you use draft mode, which I’d recommend. Nothing goes out until you approve it, and you can pause, stop or delete a campaign at any point.
How does it avoid bounces?
It verifies each address with Hunter before sending, and if a campaign starts bouncing too often it stops rather than continuing to damage your sending reputation.
Why does it skip big companies?
Because they almost never reply to cold outreach. Filtering them out improves both your deliverability and your reply rate, and saves you sending volume.
Do I need an expensive model to run it?
No. It works on subscriptions you already have, and cheaper options like a Grok or MiniMax coding plan are fine. You can also use an OAuth login.
Is this finished software?
No — it’s experimental and I’m still evolving it. It works and I use it, but treat it as a system you’ll keep improving rather than a polished product.
The bottom line
The Hermes email skill takes the six things that make cold outreach work — targeting, enrichment, verification, filtering, sending and tracking — and puts them behind one button inside your agent OS. Run it from a separate inbox, keep draft mode on, be honest that it’s an agent, and it becomes the least painful part of link building.
About Julian Goldie
I run Goldie Agency, a 7-figure SEO agency, and teach this stuff daily on a 394K+ subscriber YouTube channel. I’ve delivered 240+ client projects on Upwork at a 100% job-success score over 10+ years of ranking sites through every major Google update. The systems I actually run are inside the AI Profit Boardroom, and my link building book is free here.
