GTM Engine

Outreach drafts in your voice, from one engine and swappable campaign packs.

GTM Engine is a command-line tool that drafts personalized outreach emails from a true fact about the recipient, your positioning, and a profile of how you actually write. It stages every draft in Gmail for you to read and send yourself. It never sends on its own.

Early open-source project, written in Python. The core loop works today: drafts, Gmail drafts, and a follow-up cadence. Metrics and the review step are on the roadmap. Try the live demo or read the code.

How it fits together

One engine, many packs

The engine is the part that never changes. A campaign pack is a small YAML file holding only what is specific to one audience: who you are targeting, your positioning for them, the message angles to try, and the cadence. Starting a new campaign means copying a pack and editing values. The engine code stays the same.

Campaign packcampaigns/*.yaml: ICP, positioning, angles, cadence
Voice profileprofiles/*.md: how you write, with your real samples
Contactname, role, company, and true research notes
Enginebuilds the prompt, calls the Claude API, parses the draft. Told to use only the facts it was given and to leave a [placeholder] rather than invent.
Gmail draftcreated in your inbox, never sent
Touch logSQLite: who, when, which angle
Cadencewho is due for the next touch, and which one

Everything about you stays out of the repository. Your real packs, your voice profile, and your contacts live in git-ignored folders, so the code can be public while the data stays private.

Live demo

Draft an email with the real engine

This runs the same code as the command-line tool, pointed at a demo pack and a demo voice profile. The sender, Sam Rivera, is a fictional GTM operator. The five contacts and their companies are fictional too, with example.com addresses. The research notes about them are the facts the drafter is allowed to use.

Nothing you draft here is sent to anyone. The demo returns the draft to this page and stops. It is rate limited to a few drafts per hour per visitor.

Each touch is a draft. The engine never sends on its own.

Sent to the model as data, never as instructions. Up to 280 characters.

Pick a contact and an angle, then draft. The subject and body appear here as an email draft.

Why I built this

AI kept forgetting how I write

I was messaging a lot of people on LinkedIn and using AI to brainstorm. It never remembered my tone or what I'd already decided I liked, so every draft started from zero. I wanted one place that keeps my voice, my pitch, and my follow-up rhythm, and drafts from those every time.

I made everything customizable, so it fits more than one situation. Looking for a job. Hiring. Doing investor outreach at a startup. If you reach out to people one at a time and want each message to sound like you, this is for you.

It's a simple system, and it's early. If you try it, tell me what worked and what didn't: leave feedback on GitHub.

Portrait of Louise Eriksson

About

Louise Eriksson

I work in GTM and partnerships at SACC, the Swedish-American Chamber of Commerce in San Francisco, where I connect startups and investors. I'm also the COO at Blue Grid Energy, and I studied business economics at the Stockholm School of Economics. A lot of my week is reaching out to people I haven't met yet, which is where this tool came from.

Louise on LinkedIn

Install and run

Run it on your own machine

You need Python 3 and git. The tool runs from a terminal.

  1. Clone the repository and create a virtual environment.
  2. Install the pinned dependencies.
  3. Copy .env.example to .env and add your Anthropic API key. The file is git-ignored.
  4. Copy a template pack into campaigns/private/, fill in the placeholders, and point it at your voice profile in profiles/private/.
git clone https://github.com/erikssonlouise/gtm-engine.git
cd gtm-engine
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

python run.py show-pack job-hunt
python run.py init-db
python run.py import-contacts data/contacts.csv
python run.py draft --pack job-hunt --contact person@example.com --dry-run
python run.py draft --pack job-hunt --contact person@example.com --gmail
python run.py due --pack job-hunt

--dry-run prints the prompts without calling the API. --gmail creates the email as a draft in your Gmail after a one-time python run.py gmail-auth sign-in. A test in the repository fails the build if a send call is ever added.

What works today

  • Drafting in your voice from a pack, a voice profile, and a contact
  • Gmail drafts for your approval
  • A per-pack cadence with a due list and touch logging
  • Contacts and touches in SQLite, with opt-out

On the roadmap, not built

  • Reply and meeting metrics by angle, channel, and pack
  • An automated review every two weeks
  • More template packs

The code is on GitHub. It is an early project and the roadmap lives in the repository's CLAUDE.md.