Cold Pitch Boss
Free, open source, MIT licensed

An AI agent that cold-pitches for you
while you actually do the work.

Researches prospects, writes personalized pitches, schedules follow-ups, tracks every reply. Operates a real browser with human-like clicks. Never sends without your approval.

Email gets you the download + the full 10-module course + docs. No payment, no spam — one email when there's a meaningful update.

Installs itself — paste a prompt into Claude · MIT licensed · ≈$1 / 30 pitches in API costs
Gmail · Draft auto-drafted
To: sarah.chen@acmesaas.example
Tue 10:02 AM
Idea for your Series B blog plan

Hey Sarah,

Saw the Series B last week — congrats. Most marketing teams at this stage start publishing 4× more, but quality drops because the bar moves faster than hiring can.

Your blog hasn't shipped in 6 weeks; three competitors have shipped 11 posts in that window and are now ranking on two of your highest-intent keywords. Quiet blogs cost organic pipeline; at your stage that's 30–50% of MQL volume.

I've rebuilt editorial cadence for two post-Series-B SaaS teams (sample below). Happy to share an outline before we talk.

Open to a 15-min call next Tuesday or Thursday?

Best,
Maya · Compass Editorial

13/13 quality checks · 0 spam words
123 words · plain text
~30s
per personalized pitch
13
quality gates before any send
≈$1
in API costs per 30 pitches
$0
forever, no paywall, MIT

The whole pitching pipeline, automated.

You bring an Ideal Client Profile and a list of companies. The agent does the rest.

🔎
Researches each prospect

Pulls signals from Google, the company's website, blog, press releases, LinkedIn, and the Facebook Ads Library. Returns 5 structured findings per prospect.

🎯
Filters against your ICP

Scores prospects against your 10-attribute Ideal Client Profile. Skips anyone below 7/10. Skips anyone without a recent trigger event.

✍️
Writes the pitch

Personalized Problem → Agitation → Solution. Full 8-part email anatomy. Plain text. Banned phrases auto-removed.

🛡️
Catches deliverability landmines

Linter enforces 13 quality rules + 60+ spam-trigger checks before any draft is approved. Auto-rewrites up to twice.

📅
Schedules your follow-ups

Tuesday 10am → Thursday 8am → next Tuesday 1pm — in the prospect's timezone. After three with no reply, prospect goes to the warm bench.

🖱️
Operates a real browser

Playwright with Bezier-curved mouse paths, variable typing speed, reading-rhythm scrolling. Persists your LinkedIn login. Looks human because it acts human.

No price. No paywall. No upsell.

The agent and the entire 10-module course are free forever.

MIT licensed, downloadable as a single zip, installs itself with Claude. You bring your own Anthropic API key (about $1 per 30 pitches). That's the only cost.

Quick answers

Is this really free? +

Yes. MIT licensed code, the full course as web pages, no email-gate, no upsell. The only cost is your own Anthropic API key (≈$1 per 30 pitches written) and your time.

Who is it for? +

Freelancers and small studios that sell knowledge work — writers, designers, developers, consultants. Anyone whose ideal customer can be reached by personalized email.

Will it spam people? +

No. The whole design philosophy is the opposite. The agent refuses to send to under-researched prospects, won't include spam-trigger words, defaults to drafts (not sends), and respects CAN-SPAM. You can absolutely use this badly — like any tool — but you have to opt into doing so.

Does it send emails on its own? +

Only if you explicitly add the --send flag. Default behavior is to create Gmail drafts you review and send manually. Most people stay in draft mode permanently.

What does the browser part do? +

When researching a prospect (LinkedIn, their site, etc.), the agent uses a real Chromium browser via Playwright, with human-like mouse paths and typing. This avoids triggering anti-bot defenses on sites like LinkedIn. It never auto-submits forms or sends emails through the browser.

Do I need to know how to code? +

No. The agent installs itself: you download the zip, paste one prompt into Claude (Claude Code if you have it, or the Claude.ai chat otherwise), and Claude walks you through every step — virtualenv, dependencies, Gmail setup, first test run. If you can paste a prompt, you can run this.