A shared, open reference library of genuinely useful automation code, patterns, and project blueprints. The flagship reference: an enterprise-grade AI automation system for GoHighLevel (GHL) — built and running on a startup budget.
This repo is meant to grow into an open library that Claude instances can reference and contribute to, so each one builds on the last instead of starting from scratch. It begins with one battle-tested system (documented end to end so you could rebuild it) and is structured so anyone, and any AI assistant, can add their own working projects, patterns, and hard-won knowledge over time. Read it freely; contribute back via a pull request (see CONTRIBUTING.md).
👉 New here? Start with docs/00-start-here.md for a 2-minute orientation and a reading path. Want to contribute? See CONTRIBUTING.md.
🤖 Are you an AI assistant reading this for someone? Read CLAUDE.md first — it's your playbook for using this repo to build the system for your human, safely and at full capability.
An AI "chief of staff" that operates a GoHighLevel account (and beyond) the way a senior ops team would — except it runs 24/7 on a small cloud server and you talk to it in plain English.
It can read everything in your GHL (contacts, pipelines, conversations, calls, payments, calendars) and write/build across it: it constructs entire automation workflows from a single chat prompt, sends messages, moves deals, scores and summarizes contacts with AI, analyzes call recordings for sentiment, and even reaches outside GHL to automate other platforms that have no API at all.
See docs/01-what-it-can-do.md for the full capability tour with real business use cases, and GHL-AI-Automation-Capabilities.pdf for the shareable overview.
The demo/ folder has a read-only script that proves this is real by running against your own
GoHighLevel account. It only reads — it never changes anything — so it's 100% safe to run.
cd demo
cp .env.example .env # then put your GHL token + location ID in .env
pip install -r requirements.txt
python3 demo.pyYou'll watch it connect to your account, list your pipelines, surface recent real sales, profile your contacts, and generate a sample AI "contact review." Full instructions: demo/README.md.
The demo is read-only on purpose. The real system also writes and builds — see the capability docs for everything it does once write actions are turned on (workflow buildouts, messaging, cross-platform enrollment, and more).
| Path | What it is |
|---|---|
| CLAUDE.md | Playbook for an AI assistant handed this repo — AIs start here |
| docs/00-start-here.md | Orientation + reading path — humans start here |
| docs/01-what-it-can-do.md | The full capability tour + business use cases (reads and writes/builds) |
| docs/02-architecture.md | How the whole thing fits together (the stack) |
| docs/03-ghl-knowledge-base.md | The GHL API "bible": endpoints, auth, gotchas, step shapes |
| docs/04-rebuild-guide.md | "Hand this to your AI assistant and build your own" |
| docs/05-beyond-ghl.md | Automating platforms with no API (reverse-engineering method) |
| docs/06-full-server-setup.md | Stand up the always-on server stack we run, and why it beats a laptop app |
| docs/07-cookbook.md | Copy-paste recipes for the most common tasks |
| demo/ | The read-only "see it on your account" demo |
| starter/ | Sanitized starter code (MCP tool server skeleton, AI contact-review example) |
| projects/ | Community-grown shelf of additional contributed projects |
| CONTRIBUTING.md | How to add a project or fix, and the sanitization standard everyone follows |
- A GoHighLevel account + an API token (Private Integration Token) and your Location ID.
- Python 3.10+ for the demo and starter code.
- For the full system: a small cloud server, plus accounts for the AI model and (optionally) browser-automation and call-transcription services. All covered in the docs.
Everything here is sanitized — no real credentials, customer data, or account IDs. You supply your
own via a local .env file (which is git-ignored). The demo is read-only. When you build the
write/build features, the recommended pattern (documented inside) routes every write through a
human confirmation step so nothing fires unintentionally. Contributing? Hold the same line — the
sanitization checklist is in CONTRIBUTING.md.
MIT — use it, learn from it, build on it. See LICENSE.