[alxbridge]

A recent upgrade of the any-llm-sdk transitive dependency pulls in both the OpenAI and Anthropic SDKs, adding ~55MB to every gunicorn worker and pushing production dynos over their 512MB quota.

As the feature's not currently in use, this is removed until needed and performance is better

Tracked on TWE-712

Description of Changes Made

I've removed the changes made to add wagtail-ai, including its dependencies, configuration, and related documentation.

Once a fix can be made to django-ai-core (which is responsible for how any-llm-sdk is pulled in), this PR can be reversed and the updated versions bumped.

Verification

Memray profiling showed the following memory usage for the site code before the Wagtail 8.0 upgrade which pulled in the newer dependency, after it, and with this PR:

	Before upgrade	After upgrade	After upgrade, with this PR
Peak heap (memray)	105.1 MB	177.5 MB	106.7 MB
Worker RSS after boot	~137 MB	~205 MB	~149 MB
Modules imported	2,787	4,540	2,207
Boot time (warm)	~2.2 s	~4.4 s	~1.7 s

Deployment

wagtail-ai's migrations created two tables: wagtail_ai_prompt (populated with default prompts) and wagtail_ai_agentsettings, as well as content types and permissions in existing tables. Uninstalling the package doesn't remove any of these. To clean them up:

1. roll back the wagtail-ai migrations while the app is still installed. This drops both tables and their django_migrations rows:

heroku run -a torchbox-com-[staging/production] -- python manage.py migrate wagtail_ai zero

2. Deploy.

3. After deployment, remove the stale content types for the uninstalled app. This also deletes their permissions and any group permissions assigned to them:

heroku run -a torchbox-com-[staging/production] -- python manage.py remove_stale_contenttypes --include-stale-apps

It lists what it will delete and asks for confirmation. Check that the list only contains wagtail_ai entries before answering yes. Drop --include-stale-apps and it won't touch content types for uninstalled a
pps at all.

MR Checklist

• Add a description of your pull request and instructions for the reviewer to verify your work.
• If your pull request is for a specific ticket, link to it in the description.
• Stay on point and keep it small so the merge request can be easily reviewed.
• Tests and linting passes.

Unit tests

• Added
• Not required

Documentation

• Updated build docs
• Updated editor guidelines (See https://intranet.torchbox.com/torchbox-com-project-docs for the private link, for Torchbox employees only)
• Updated field spec (See https://intranet.torchbox.com/torchbox-com-project-docs for the private link, for Torchbox employees only)
• Not required

Browser testing

• I have tested in the following browsers and environments (edit the list as required)
  • Latest version of Chrome on mac
  • Latest version of Firefox on mac
  • Latest version of Safari on mac
  • Safari on last two versions of iOS
  • Chrome on last two versions of Android
• Not required

Data protection

• Not relevant
• This adds new sources of PII and documents it and modifies Birdbath processors accordingly

Light and dark mode

• I have tested the changes in both light and dark mode
• The change is not relevant to dark and light mode

Accessibility

• Automated WCAG 2.1 tests pass
• HTML validation passes
• Manual WCAG 2.1 tests completed
• I have tested in a screen reader
• I have tested in high-contrast mode
• Any animations removed for prefers-reduced-motion
• Not required

Sustainability

• Images are optimised and lazy-loading used where appropriate
• SVGs have been optimised
• Performance and transfer of data considered
• If JavaScript is needed alternatives have been considered
• Not required

Pattern library

• The pattern library component for this template displays correctly, and does not break parent templates
• The styleguide is updated if relevant
• Changes are not relevant the pattern library

[zerolab]

@alxbridge I say let's fully nuke it and revisit once improvements are released upstream