Mid-2026 Holy Productivity Trinity
Super-charging Life
I have a lot of interests. Although I wish I were better at cricket, that's a minor concern, amply compensated for by the pursuit of other fun things to do.
Most of my interests, it turns out, can be enhanced and accelerated by the use of AI. Many of these will show up later in our blog. This article is about the three tools I use to get things done - I hope they will work for you, too.
Github
I get nervous when I'm not sure where my projects "live". I can never use a system that promises to "keep my work" on their servers, or "you can roll back your changes by clicking the History button". Nope, not good enough. I reach for an older grande dame of safe keeping: version control.
Git is a form of "version control". I remember CVS and SVN before Git. These tools were developed for coding: keeping your precious code safe, and the full history of the changes that you made, at your fingertips and available for all time. The invention of version control probably did more for coding than the invention of any one language.
Developers could "check out" their code from their own little local library, mess with it, possibly ruin it, and if they really got into trouble just revert back to the library version. If things went well, you could check back into the library and your library would keep your latest change. In fact it would keep all changes, and you could always revert back to any historical point in your project, like using a time machine to get yourself out of trouble.
Version control keeps a local "repository", a.k.a. library or vault, and you can also push your code to a remote library ("repo"), for example on a site like GitHub. That's what most people do today.
Version control is not just for code; you can use it for any project, although it works best for text-based projects. Large files like video need slightly different tricks, but it all still works.
Perhaps best of all - you can check in your project from your laptop, say, and check it out again on your PC. So, it makes your life portable. I work like this all the time and it's great to check in here, check out there, banking saved checkpoints as you go.
Therefore using GitHub is my first recommendation: keep your projects and their history safe and portable!
How to use Git and GitHub
You need a GitHub account and a copy of Git on your computer. Both are free for ordinary use. Sign up at GitHub, install Git, make a new repository for a project, and clone it onto your machine. The GitHub Desktop app is a perfectly respectable place to start if the command line feels like an unnecessary test of character.
The commands worth knowing are not many. git status tells you what has changed. git add selects the changes you want to save. git commit -m "a useful note" makes the local checkpoint. git push sends it to GitHub; git pull brings changes back down from GitHub. That is most of it. You can learn the rest precisely when you need it, which is how I prefer to learn most things.
Hermes
GitHub is your project hub, but your engine of creation could be Hermes Agent. Hermes is similar to Claude but it's open source and free to use. Well, it needs an LLM engine to drive it, but Hermes provides things like agents, memory, skills and basically works for you 24/7. You can set it up so you can WhatsApp it instructions, but there are potential security issues and I avoid that.
What is Hermes? It's a program you install; you run it in a terminal (in Windows, WSL works well for this), or on Linux or Mac. It asks you for some setup which is mostly unimportant. Just one thing really matters, and that is which model to use as its brain. You can choose some free options here, give it your API keys to an OpenAI or other model account, or, my favourite: an API key for an OpenRouter account.
Then what?
Hermes is a general-purpose thing. You can ask it "I would like you to submit my tax self assessment to the Government. What do we need to do?" This is clearly amazing magic that we all need. It actually works. Hermes will do the research it needs to do, ask you questions, perform actions for you, ask you for permission and generally get the job done.

Safety
Should you trust it with personal data like your income and tax details? I'm going to wash my hands here and say that's entirely up to you. Your risk; don't blame me if this goes wrong. It seems to make an attempt to keep your data private, but let's face it - your financial details are possibly leaking out onto the internet and almost certainly leaking into the LLM provider.
Please take this into account and just don't trust Hermes or your LLM provider with any data you cannot afford to leak
Look at the possibilities ...
If you don't want Hermes to do your taxes for you, well, ask it to code an app for you, or write a book for you, or use any MCP server or any skill from the internet to do basically anything at all you can think of. Some examples I have tried so far:
- Creative writing
- Setting up this blog
- Writing a game in flutter
- Writing another game in JavaScript for web
- Finding the best place to live in the country, taking many available public data sources into account
I have not run into something it has failed at, so far.
How to install and set up Hermes
The current easy route on Mac or Windows is the Hermes Desktop installer from the Hermes website. If you want the terminal-only version, Mac, Linux and WSL use the little installer command on the Hermes Quickstart page. Native Windows has its own PowerShell installer. I would use WSL on Windows if you are comfortable with it: it makes the internet's endless Linux-shaped instructions much less irritating.
When it is installed, run hermes setup or hermes model, choose a provider and a model, then make it answer one normal question before configuring every clever extra. Hermes can use a browser when you give it the relevant tools, but start simply. A terminal, a model and a small project are enough to find out whether this changes how you work.
OpenRouter - the model gateway
The missing piece is access to an LLM brain. This used to be an OpenAI or Anthropic account, but it's pretty clear that some alternative models are available at a fraction of the price and are only a few months behind the US-based frontier models. I did not want to sign up for ten different accounts, including sharing credit card details with multiple foreign companies. Instead, I signed up once with OpenRouter. OpenRouter gives you one account, with one API Key, and allows you to route to an LLM provider via them. Most of the models you could possibly want, from any country, whether large or small, cheap or expensive, are available right there on OpenRouter. Brand new models, by the time you read about them, are often already available.
So the advantages are:
- Model choice
- New models are made available very quickly
- Auto options, for example routing to cheap models, or the "auto beta" option that routes to whatever model is popular - quite clever
- One account instead of many
- Pay per use, you pay the same token rate as you would directly with the model provider.
The catch
OpenRouter needs to make money but their profit margin is slim and transparent: when you top up your credits, you pay them an explicit fee, a small percentage on top of the credits that you purchased. I find this to be perfectly reasonable at time of writing, I'm happy to pay it and it is fully transparent.

How to sign up and use OpenRouter
Go to OpenRouter, make an account and add a small amount of credit. Then create an API key in its keys section. Treat that key like a password: it buys things, and it should not end up in a GitHub repository, a screenshot, or an article about productivity.
Back in Hermes, run hermes model, choose OpenRouter and paste in the key when asked. You can change models later without creating another account. Start with a modest budget and watch the usage page for a week. It is much more relaxing to discover your habits on a few pounds than on a surprise invoice.
Conclusion
Our holy (unholy?) trinity of productivity daemons has opened up a new world of things I can do, that I would simply not have time for if doing it without AI. It's not that AI in my case is taking work from humans - these projects would not have existed without AI.
More on these projects in upcoming posts - good luck with your own adventures.