Launching today

Megamesh.sh
Your own self-hosted WireGuard VPN, with one-click deploy
20 followers
Your own self-hosted WireGuard VPN, with one-click deploy
20 followers
Your own VPN, self-hosted and built on WireGuard. Install it on a small VPS with a one-click deployer, a readable script or an AI-agent prompt. Free, open source, MIT. A personal VPN, not an anonymity service: read the limits first.







1/4 · Day 7 of 11 of the Product Hunt Marathon: 11 free, open-source products, one launch a day, 10 to 20 November.
This one comes from my cybersecurity side: I've worked in the field and still do from time to time.
Megamesh is your own VPN. It's self-hosted, built on WireGuard, and installs with a one-click deploy on a small VPS you control. You get a hardened personal VPN with a web admin page, and you add a phone or laptop by scanning a QR code.
Why run your own? Renting a tunnel, you're a subscriber: shared servers, a privacy policy you take on faith, a box you'll never see. Running one, you're the operator, and every line of the setup is on GitHub. I didn't invent a protocol. WireGuard, wg-easy and Caddy do the heavy lifting and deserve the credit; my part is the setup glue.
Read the limits before you run it. It's a personal VPN, not an anonymity service, a privacy guarantee or a security certification. It hides your traffic from the café Wi-Fi and your ISP; your VPS provider still sees it leave its network. One operator, not a multi-tenant platform, and owning the server means owning its updates, keys, recovery and costs.
What would you want checked before trusting a VPN on your own server?
2/4 · Three ways to install, and what each asks you to trust
One-click deploy (https://megamesh.sh/deploy): four things, then wait. Give it a server IP, root password, domain and email. It checks the server, shows the plan, installs in 10 to 30 minutes, then hands over your admin login and the one DNS record to add. You trust it with your root password, which transits megamesh.sh during the install and isn't stored.
One Bash script: readable in about ten minutes, run as root on a fresh VPS. You trust what you've read.
PROMPT.md: a documented prompt for an AI agent that can run commands over SSH. You trust the agent with that access.
All three lead to the same setup: WireGuard for the tunnel, wg-easy for devices, QR codes and stats, Caddy for automatic Let's Encrypt TLS on the admin page, a default-DROP firewall with three ports open (SSH tcp/22, WireGuard udp/51820, HTTPS tcp/443), fail2ban, daily OS updates and weekly health-checked upgrades that roll back on failure.
3/4 · Limits, licences and the code
The limits belong in the launch, not the small print. The deployer leaves SSH on password login until you add a public key, and 2FA and VPN-only admin mode are yours to switch on. CentOS Stream 9 is the reference server and Ubuntu 24.04 is verified end to end, deployer included. For Rocky, Alma or Debian, check the README's test notes first. Rollback code isn't proof that every failed update recovers safely.
No account with Megamesh, nothing stored, nothing logged. Before Megamesh, the reference server took 76,073 failed SSH logins in 25 days.
The project code is MIT: use it, extend it or commercialise your own work while keeping the required notices. Upstream projects keep their own licences, so read them before redistributing a combined product.
Start here: https://megamesh.sh
Read the code: https://github.com/paulfxyz/megamesh
Follow the experiment: https://ph.paulfleury.com/
4/4 · Take it further, then what comes next
Yours to take further: give Astra or your preferred AI the repository, one narrow goal and a budget. Ask for a plan, tests and a security review, not a blind root command. Iterate with human review, then bring back a focused PR or a sanitised issue. Please don't post secrets or private server details.
I'll review the feedback, prioritise bugs and useful small improvements, and do my best to fund the AI work needed to iterate. That isn't a promise to implement every request.
Tomorrow, Day 8 of 11: AI Credit Checker. The bigger CHO idea is not "let AI run everything"; it's making someone accountable for turning real needs into tools the organisation can understand and maintain.
Separate from the marathon, a few ventures of mine, none of them launching on Product Hunt in it. Openline (openline.com) is my main venture and startup: it aims to reshape the eSIM market, with a deep pipeline of ideas ready to execute, and I'm raising, so I'm looking for VCs, angels and investors. Iris (iris.online), one inbox for customer messages across 19 channels, and Digital Spot (digital.spot), social media on autopilot, are spin-offs of tools built for Openline; for both I'm looking for investors and early customers. Enki (enki.ngo) is rather a thought experiment: a non-profit, local-first AI public good, too big for me alone, where I'm looking for philanthropists, donors and helpers. And Cup of News (cupof.news) is a Sunday-afternoon hobby that is soon turning into an app, an open-source, self-hosted AI morning digest; contributors and users are welcome.
A playful title for a serious role: I'm open to a CHO (Chief Hacking Officer) brief of up to 30h/week with a great company or startup, if the product is worthwhile, there is trust and a modest budget.
Web Summit Lisbon is over, but I live in Lisbon and I'm happy to meet around the city. Thank you to everyone I met there. You can reach me by DM on X (@paulfxyz) or LinkedIn, or at paulfleury.com/contact.