I thought I’d share my successes with using chatGTP to perform bulk edits in Tiddlywiki tiddlers and their content
Central Street Archive
In this TW, I wanted to remove all tags that were street addresses. The job looked very daunting given over 2000 tiddlers, many have one or more tags as street addresses.
I exported all my tiddlers into a json file, uploaded this to ChatGPT and asked it to identify all tags that were a street address and return the result in a CSV file.It did this and I checked the file and it looked OK. So I asked it to do it again but this time remove the street address tags and return the edited file in JSON format.
It did that and I imported the edited JSON file back into my TW and checked the tiddlers listed in the CSV file with those in the edited TW. All seems OK.
Time Taken: 5 minutes.
Gun Wiki
In this TW I am trying to identify the names of all guns, ships, forts and manufacturers and wikifi them by enclosing them in double square brackets and for gun names, suffix them with “|gtname” where gt name is the code for the gun type held in a match file (gun name,GT-254)
my recommendation for tasks like this is to let the AI generate a script for you, that performs the requested task. I would execute this script locally, hence the AI never sees your tiddlers with your data and maybe your personal information.
In addition, this allows you to review the performed tasks in the script code, so the tasks is not so much a black box, and you may learn something.
Kind regards,
Eike
I am not an chatGPT user, hence I am not aware of any limitations regarding this type of task, but expect it to be able to generate scripts and code, because somehow it’s text, too.
How I do it: VSCode Copilot + TidGi Desktop , side by side
Deep topic use GPT5.5 / Claude Opus 4.7, copilot allow config custom provider, and its edit to .tid files on wiki folder will immediately reflect on TidGi Desktop. (need to open Expermental filesystem watch on workspace setting.)
Simple task like Batch editing files, you can use deepseek, is very cheap. And if it found mission size is very large, it will automatically write a shell script or Python script to update them.
An example here, I tell the “silly model” deepseek v4 flash to change name conversion of wiki, it automatically decide to use PowerShell. deepseek v4 flash is the dumbest model in my daily model list, it still performs good.
Don’t let deepseek v4 flash run any shell command!
I find it delete all metadata and argue about this with it, it says:
This is a major problem! There is a serious bug in my PowerShell script: When it is replacing English terms, the regular expression (type: text/vnd\.tiddlywiki\n)(.+) only matches the text starting from “type:”, but when writing, it only concatenates “type:” + the text, losing the entire block of meta - information (such as “title:”, “created:”, “tags:”, etc.) in front of “type:”.
Thanks everyone. I’m continuing my use of ChatGTP and find that its suggestions for filters is not too bad, at least for my simple ones at any rate. Also its bulk conversion is good, even better than a repeated BBedit run
