I Built a 9-Agent AI System That Runs My Cold Outreach. Here's Exactly How.

After my video went up, everyone asked the same thing: "Can you send me the n8n file?"

So here's the whole thing — in simple words. What it does, how it works, and how you can build your own. No fluff.

What it does (in plain English)

Every morning, before I even open my laptop, this system has already:

• Found new companies that are a good fit for me
• Found the right person at each company — and their real, working email
• Written a personal email to each one, mentioning something true about their business
• Checked every email for quality, so nothing weak reaches me
• Put them all in one email for me to review

I open that email on my phone, and for each one I tap Send, Edit, or Skip. Takes about 10 minutes. That's all I do.

After that, the system keeps working on its own:

• Someone replies? It stops all follow-ups to them right away.
• Email bounces? It finds a new working contact at that company.
• No reply? It follows up politely — on day 3, day 7, and day 14 — in the same email thread.
• Someone interested? It moves them to a special list so they get more attention.

The one rule I never break: nothing gets sent without my approval. I write zero emails. I just approve. That's the trade.

How it works: the 9 helpers

Think of it as 9 small robots, each with one job. Each one does its task, saves the result, and passes it to the next.

The full details for each one are in documentation.md.

3 simple lessons (that will save you time)

1. Put a spending limit on anything that searches the web. Some "free" web-search tools get expensive fast when a robot uses them all day. I use a cheaper dedicated tool (Tavily) with a hard limit. Always cap your spend before you let it run on its own.

2. Let AI be creative in a small box. Let code handle the rest. If you ask AI to remember 20 rules every time, it will forget one. So I only ask the AI for the creative bits (the first line, the reason it matters). Plain code handles the fixed parts (greeting, subject, sign-off) — the same way every time. Fewer mistakes.

3. Check every email address before you send. This system checks each email at three points: when it's found, before the review email, and again the moment you tap Send. Sending to a dead address hurts your reputation. A quick check avoids that.

How to build your own (step by step)

You don't need to be a coder. Follow this order:

1. Read documentation.md to understand what each helper does.
2. Follow BUILD-YOUR-OWN.md — it walks you through everything: the database, the accounts, and the setup.
3. Import the 9 files from the workflows folder into your own n8n. (All keys are removed — you add your own.)
4. Add your own keys for search, email-checking, and the AI.
5. Rewrite the prompts to match your business and who you help.
6. Test small first. Run it on a few leads for a few days before you trust it fully.

What you'll need

• An n8n account (free to start)
• A free Supabase project (this is the shared memory)
• A Gmail inbox just for sending
• API keys for: Tavily (search), AbstractAPI (email check), and an AI model

All of these have free tiers. Set aside an afternoon to wire it up, and a few days of light testing before going live.

Grab the files below, follow the guides, and you'll have your own version running. If you get stuck, the guides cover the tricky parts in detail.