I suggest starting with something very simple — just to get the feel of it.
Take your Gramps XML file… and ask an AI to help you write a tiny script that extracts a few fields and writes them into a plain text format (Markdown, CSV, whatever you prefer).
And by the way — you don’t even need to start with a Gramps XML file.
You can begin with any text‑based file you already have lying around: a .txt, .md, .csv, or anything with some structure in it. The point is simply to learn how to ask the AI to read a file, extract something, and write a new file.
Once you understand that workflow, switching to the Gramps XML format is just a matter of changing the input file and adjusting the extraction logic. The process is exactly the same.
I still use the Copilot panel in Microsoft Edge for this kind of thing, and I edit and run the code in Visual Studio Code. You can do the same, or just use Google AI in a browser tab until you get comfortable.
The key is to speak plainly in the prompt. For example:
“Can you write a complete Python script that I can copy into Visual Studio Code, which reads the Gramps XML file named ‘[actual filename]’, extracts the full name and birth date for all people, and writes them to a Markdown file called
people.md?
Format the birth date as ‘dd. mmm. yyyy’.
Also explain how to install Python and how to run the script in the terminal.”
And one important thing:
If the AI gives you code that doesn’t run, or doesn’t do exactly what you want, don’t give up.
Just copy the error message from the terminal and paste it back to the AI. Or tell it:
“The script didn’t do what I wanted. Here is the exact output I got: [paste].
I want the result to look like this instead: [describe the format in your own words].”
After a few iterations, the AI will understand your instructions better and adjust the code.
This is completely normal — you almost always need two or three rounds before the script does exactly what you want.
And finally:
Always tell the AI that this is about the Gramps genealogy software project.
There are several unrelated software projects called “Gramps”, and if you don’t specify, the AI may confuse them. Just add a line like:
“This is for the Gramps genealogy software XML format.”
Once you’ve done one or two of these small tasks, everything else becomes much easier.
Once you start getting comfortable with giving the AI clear instructions — and the small scripts you “architect” actually run — you can take the next step and ask the AI to help you build a Gramplet. This is much easier than it sounds.
Even the Copilot panel in a browser is surprisingly good at:
-
setting up the correct file structure
-
generating the Python code
-
adding the right imports and dependencies
-
explaining where each file should go
-
and making sure the code matches the Gramps genealogy software API
Just make sure you always tell the AI:
“This is for the Gramps genealogy software project.”
There are several unrelated software projects called “Gramps”, and the AI may otherwise confuse them.
A nice thing about Gramplets is that they don’t need to be included in Gramps itself. You can keep them as local add‑ons. And the AI knows how to register a Gramplet so it appears exactly where you want it — sidebar, dashboard, person view, whatever.
Once you’ve done a few small scripts, asking the AI to build a Gramplet becomes a natural next step. You stay the architect, and the AI does the heavy lifting.
PS. One more thing that will save you a lot of frustration later:
save your conversations with the AI.
Just copy everything out and store it as a plain text file or a Markdown file. If you don’t already use a note system, that’s totally fine — but having a record of what you tried, what worked, and what didn’t is incredibly useful.
I use Obsidian and the FOAM extension in Visual Studio Code for this kind of technical note‑taking, but you can use anything you like.
Why this matters:
-
You can go back and see exactly what instructions produced working code.
-
You can copy the whole conversation back into the AI if your chat gets reset or the browser closes.
-
You can build your own little “playbook” of prompts and working scripts.
-
You avoid losing good solutions because of a refresh or a timeout.
Most AI clients save chats automatically, but not all of them — and not always reliably. Having your own copy means you’re never dependent on the UI.