You can build a self-contained email outreach engine with Hermes Agent — the open-source AI agent by Nous Research — that finds prospects, enriches them with real contact names, verifies their email addresses, drafts personalized messages, and waits for your approval before sending anything. No expensive SaaS subscription. No copy-pasting between ten tabs. The whole pipeline runs from one terminal or messaging surface, and with draft mode enabled, you keep full control over what goes out under your name.
Last verified: 2026-08-06
- Hermes Agent is free, MIT-licensed, and runs on any $5/month VPS or your laptop
- The pipeline uses three tools: Hunter API for email finding/verification, Firecrawl for web scraping, and Himalaya CLI for terminal email
- Draft mode is the single setting that prevents embarrassing sends
- Average B2B cold email reply rates have fallen to 3.4–5.8% in 2025–2026 — targeting quality matters more than volume
What Is Hermes Agent and Why Use It for Email Automation?
Hermes Agent is an open-source, self-improving AI agent built by Nous Research and released under the MIT license. Unlike a chatbot that only answers questions, Hermes lives on your server, remembers what it learns across sessions, and can actually execute tasks — web searches, browser automation, file operations, code execution, and sending messages through its built-in email gateway. It connects to Telegram, Discord, Slack, WhatsApp, Signal, and Email, so you can trigger workflows from wherever you are.
For email outreach specifically, Hermes matters because it bridges the gap between "AI that writes copy" and "AI that runs the entire pipeline." Most cold email tools handle one piece — finding leads, or writing copy, or sending. Hermes orchestrates all of them in a single agent loop, with persistent memory so it learns which prospects respond and which don't.
If you're new to Hermes, our guide to running Hermes Agent for free covers the zero-cost setup path, and the Hermes Agent power user playbook walks through the settings that matter most.
How Does Email Automation With Hermes Agent Actually Work?
The pipeline works in six sequential steps — each one quietly removing a reason an email might fail. You describe who you want to reach, the agent enriches each company into a real person with a real email, verifies that email works, filters out targets too large to respond, drafts personalized messages from a dedicated inbox, and tracks replies. The critical safety layer is draft mode: when it's on, nothing sends until you read and approve each message.
Here's the architecture at a glance:
| Step | What Happens | Tool | Key Detail |
|---|---|---|---|
| 1. Lead description | You describe the target audience in plain language | Hermes Agent | No spreadsheets — just a prompt |
| 2. Enrichment | Companies become real people with names and roles | Hunter API + Firecrawl | Turns info@ addresses into human contacts |
| 3. Email verification | Each address is checked for deliverability | Hunter Email Verifier | 0.5 credits per verified email — no charge on failure |
| 4. Filtering | Large companies get skipped (low reply odds) | Hermes Agent logic | Giant companies rarely respond to cold email |
| 5. Drafting + sending | Messages written from a dedicated inbox | Himalaya CLI + SMTP | Draft mode: nothing sends without your approval |
| 6. Reply tracking | Inbound responses logged and surfaced | Himalaya CLI (IMAP) | Read replies inside the agent — no inbox switching |
What Tools Do You Need to Build This?
The entire stack is four connections. Three are API-based services; one is Hermes itself.
Hermes Agent (free, open source)
Install with one command: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash. It runs on Linux, macOS, and WSL2 with zero prerequisites — the installer handles Python, dependencies, and the virtual environment. Hermes supports multiple LLM providers (OpenRouter, Anthropic, OpenAI, local models), so you're not locked to one vendor. Your agent OS setup can run this alongside other workflows.
Hunter API — email finding and verification
Hunter finds a person's real email from their name and company domain, then verifies whether that address actually works. Pricing (verified 2026-08-06):
| Plan | Monthly Price | Credits/mo | What Credits Buy |
|---|---|---|---|
| Free | $0 | 50 | 50 emails found or 100 verified |
| Starter | $49 | 2,000 | ~$0.025 per email found |
| Growth | $149 | 10,000 | ~$0.015 per email found |
| Scale | $299 | 25,000 | ~$0.012 per email found |
One credit = one email found via Domain Search or Email Finder. Verification costs 0.5 credits per email. If no email is found or verification fails, no credits are charged. Annual billing saves 30%.
Firecrawl — web scraping for lead research
Firecrawl pulls content off websites so the agent can research prospects before drafting messages. Pricing (verified 2026-08-06):
| Plan | Price/mo | Credits/mo | Cost per scraped page |
|---|---|---|---|
| Free | $0 | 1,000 | $0 (free tier, one-time grant) |
| Hobby | $16 | 5,000 | $0.003 |
| Standard | $83 | 100,000 | $0.0008 |
| Growth | $333 | 500,000 | $0.0007 |
One credit = one page scraped or crawled. The free tier gives 1,000 credits as a one-time grant (not monthly), with 2 concurrent requests.
Himalaya CLI — terminal email client
Himalaya is a command-line email client that handles IMAP and SMTP. Hermes uses it as a bundled skill to read, compose, send, and search email directly from the terminal — no GUI email client needed. Install it with curl -sSL https://raw.githubusercontent.com/pimalaya/himalaya/master/install.sh | PREFIX=~/.local sh or brew install himalaya on macOS.
How to Set Up the Email Outreach Pipeline: Step by Step
Step 1: Create a dedicated agent inbox
Use a separate email address exclusively for this agent — never your personal or work inbox. Two reasons: (a) it keeps the agent's sends isolated from anything important, and (b) it's honest. People should know they're hearing from an AI agent, not a human pretending to be one. State this in the email body itself.
Configure the inbox in Hermes by adding IMAP/SMTP credentials to ~/.hermes/.env:
EMAIL_IMAP_HOST=imap.gmail.com
EMAIL_SMTP_HOST=smtp.gmail.com
EMAIL_IMAP_PORT=993
EMAIL_SMTP_PORT=587
EMAIL_PASSWORD=your-app-password
EMAIL_ALLOWED_USERS=your@email.com
EMAIL_POLL_INTERVAL=15
If you use Gmail with 2FA, generate an app password — your regular password won't work.
Step 2: Connect Hunter and Firecrawl APIs
Add your Hunter API key and Firecrawl API key to the Hermes environment:
export HUNTER_API_KEY=your-hunter-key
export FIRECRAWL_API_KEY=your-firecrawl-key
Hunter offers 50 free credits on the free plan — enough to find and verify roughly 33 emails before you need to upgrade. Firecrawl's 1,000 free credits let you scrape 1,000 pages for lead research.
Step 3: Describe your target audience
Tell Hermes who you want to reach in plain language. Be specific: describe the type of person, the kinds of companies, and what they do.
Example prompt to Hermes:
"Find 20 people who run AI-focused newsletters with under 50 employees. For each, I need: the person's name, their role, the company website, and a verified email address. Skip any company with more than 200 employees."
The agent uses Firecrawl to search and scrape relevant sites, then uses Hunter's Domain Search to find email addresses at those domains.
Step 4: Enrich — turn companies into people
The enrichment step is the difference between an email that gets ignored and one that gets read. Instead of sending to info@company.com, Hunter's Email Finder takes a person's name and domain and returns their direct email. The agent can also use Firecrawl to scrape the company's About or Team page to find the right person to contact.
This step is where most cold email fails. Research-based personalization lifts reply rates by up to 142% compared to generic outreach. The agent does the research; you benefit from the lift.
Step 5: Verify every email before sending
Hunter's Email Verifier checks whether an address is deliverable — it looks at the domain's SMTP configuration, the mailbox's existence, and whether the address accepts mail. This costs 0.5 credits per email (half what finding costs). If verification fails, no credit is charged.
Why this matters: Google's bulk sender requirements (enforced since 2024 for anyone sending 5,000+ emails/day) require that user-reported spam rates stay below 0.3% — and hitting a dead mailbox is one of the fastest ways to tank your deliverability. Verified emails keep bounce rates low and your sender reputation intact.
Step 6: Enable draft mode — the safety net
This is the most important setting. With draft mode on, the agent writes every email, lines them all up, and then waits. Nothing sends until you read each message, edit or delete what looks off, and give explicit approval. Only then does anything go out.
The agent also auto-stops if an email keeps bouncing. You can pause, stop, or delete a campaign at any point, and read replies directly inside the tool.
What Are the Cold Email Benchmarks You Should Know in 2026?
Setting realistic expectations matters — this pipeline doesn't magic away the fundamentals. Here's the data:
| Metric | 2023 | 2025–2026 | Source |
|---|---|---|---|
| Average B2B reply rate | 6.8% | 3.4–5.8% | Instantly Benchmark Report 2026, Belkins 2025 study |
| Target for "good" campaigns | ~7% | 5%+ | Empra Labs |
| Top decile reply rate | — | 10.7%+ | Instantly platform data |
| Emails to land one meeting | — | ~344 | Gong, 2025 |
| Google spam rate threshold | — | <0.3% (hard cap) | Gmail bulk sender guidelines |
Key takeaway: targeting quality beats volume. Belkins' 2025 study found that campaigns sending to 50 or fewer recipients had a 5.8% reply rate, while blasts of 500+ collapsed to 2.1%. One contact per company replied at 7.8%; ten or more per company dropped to 3.8%.
How to Keep Deliverability Safe When Automating Email
Email automation fails silently when deliverability breaks. Here's what to watch:
- Authenticate your domain. Set up SPF, DKIM, and DMARC records. Google and Yahoo require all three for bulk senders (5,000+ emails/day) as of February 2024.
- Keep spam complaints under 0.3%. Google's Postmaster Tools show your complaint rate. Anything above 0.1% is a warning zone; 0.3% makes you ineligible for mitigation.
- Support one-click unsubscribe. Required by Gmail and Yahoo for bulk senders (RFC 8058).
- Use a dedicated sending domain. Don't risk your primary domain's reputation. Set up a subdomain like
outreach.yourdomain.comwith its own SPF/DKIM records. - Warm up the inbox. Start with 10–20 emails/day and ramp up over 2–4 weeks. A cold inbox sending 200 emails on day one will land in spam.
- Verify before sending. Every email in the pipeline goes through Hunter's verifier. A bounce rate above 2% signals trouble; above 5% can trigger automatic filtering.
What This Means for You
If you're a solo builder, freelancer, or small team doing outreach, the Hermes-led pipeline replaces a $200+/month SaaS stack with a self-hosted agent that costs nothing beyond API credits. The key advantages:
- Total control: Draft mode means nothing goes out without your explicit approval. You catch errors, tweak tone, and delete anything that feels off — before it's sent.
- Lower cost: Hunter's free plan (50 credits) plus Firecrawl's free tier (1,000 credits) lets you test the entire pipeline for $0. Scaling to 2,000 emails/month costs $49 for Hunter — that's the ceiling for most small operations.
- Honesty as a feature: Disclosing that the outreach comes from an AI agent builds trust, and trust gets replies. The alternative — pretending a bot is a human — is how you lose sender reputation and get reported as spam.
- Skill portability: The Hermes email skill is a SKILL.md file you can version, share, and improve. Your outreach workflow becomes a reusable artifact, not a locked-in SaaS configuration.
If you're already running parallel AI workstreams, the email outreach pipeline joins them as one more tab in your agent OS — talking to the same memory, the same skills, and the same model.
Tips for Getting Started
- Always start in draft mode. Read the first batch yourself before trusting the agent to run unsupervised.
- Use a fresh inbox, not your main one. Keep them separate — for safety and for deliverability reputation.
- Be honest in the message. Tell people an AI agent is reaching out. It builds trust, not suspicion.
- Keep targeting tight. A small, well-matched list of 20–50 people beats a random list of 500 every time — the data backs this up.
- Start small. Send a handful. See what comes back. Grow from there. Simple beats fancy.
FAQ
Q: Is cold email still effective in 2026?
A: Yes — when done with verified addresses, real personalization, and honest messaging. The average B2B reply rate has fallen to 3.4–5.8% (from 6.8% in 2023), but top campaigns still clear 10%+ through tight targeting and good deliverability. The problem was never email itself; it was lazy, spammy email.
Q: How much does this cost to set up?
A: $0 to start. Hermes Agent is free and open source (MIT). Hunter's free plan gives 50 monthly credits. Firecrawl's free tier gives 1,000 one-time credits. Scaling to 2,000 emails/month costs $49/month for Hunter. If you run Hermes on a $5/month VPS, your total monthly cost is under $60.
Q: What is draft mode and why is it important?
A: Draft mode means the agent writes every email and queues it, but nothing sends until you read and approve each message. It prevents the agent from sending something embarrassing, inaccurate, or off-brand with your name on it. It's the single most important safety setting in the pipeline.
Q: Can Hermes Agent read replies as well as send emails?
A: Yes. Through the Himalaya CLI skill (bundled with Hermes), the agent can list, read, search, and manage emails from the terminal using IMAP. This means reply tracking happens in the same tool — no switching to a separate inbox.
Q: Does this comply with Google's bulk sender requirements?
A: The pipeline supports compliance: Hunter's email verification keeps bounce rates low, draft mode prevents volume spikes, and you should set up SPF/DKIM/DMARC on your sending domain. Google requires these for anyone sending 5,000+ emails/day and enforces a spam complaint rate below 0.3%.
Q: What LLM models can Hermes Agent use?
A: Hermes is model-agnostic. You can connect it to OpenRouter, Anthropic, OpenAI, local models via Ollama, or any OpenAI-compatible API endpoint. The agent's effectiveness depends on the model you choose — better models write better email copy and do richer prospect research.
Discussion
0 comments