🇪🇸Spain @gxleano

gxleano → created an issue.

We would also need to check the VDB providers which are extending from AiVdbProviderClientBase and need to include embeddingValidator.

The code looks good to me so far Scott.

After we include this on 2.0.x we could handle https://www.drupal.org/project/ai_provider_openai/issues/3552523 ✨ Opt-in to cache embedding lookups Active

gxleano → made their first commit to this issue’s fork.

And yes, it is also not completed 😂

Thanks Scott!

I was just checking what there was done, and what we should do here.

Currently working on it.

I would say that this issue could be move to Fixed after https://www.drupal.org/project/ai/issues/3477973 📌 Dynamically load Tokenizer after selecting Embedding Engine Postponed is validated.

1. https://www.drupal.org/project/iconify_icons →
2. https://www.drupal.org/project/tagify →

gxleano → made their first commit to this issue’s fork.

Thanks @ishani patel, it is already working as expected

So, moving to RTBC.

There is a related issue with the other boost plugins which will be fixed here: https://www.drupal.org/project/search_api/issues/3550113 🐛 Undefined array key "boosts" warning in NumberFieldBoost processor on Drupal 11 Active

gxleano → created an issue.

gxleano → made their first commit to this issue’s fork.

It will be included on 1.2.1 release

I would like that someone could take a second look on this after been moved to 2.0.x

Moving this to 2.0.x because of it is not a bug fix, so it could wait.

Will do, thanks Artem for review it.

Thanks @tmiguelv to do a second test there.

Moving it to RTBC.

gxleano → made their first commit to this issue’s fork.

Yep, thank you Marcus!

So, we have added a new logic with three possible scenarios:

Title NOT in contextual content + exclude_title FALSE (default) -> Title IS auto-added
Title IS in contextual content + exclude_title FALSE -> Title NOT auto-added
Title NOT in contextual content + exclude_title TRUE -> Title NOT auto-added

So, we would add the title automatically when it is not added as contextual content and the option "exclude_title" is FALSE?

gxleano → made their first commit to this issue’s fork.

I've been reviewing changes introduced on this issue and everything seems to be working as described.

Steps to perform
- Enable raw embedding vector retrieval in the Search API server
- Reindex content and extract a known normalised vector from an indexed entity (drupal_entity_id)
- Execute a search with $query->setOption('vector_input', $extracted_vector) instead of a text query

Results
- Search has been completed successfully without triggering an extra LLM call
- Top result matched the entity from which the vector was originally retrieved

I've been testing using Ollama and OpenAI and everything works as expected.

I've also added a small change to fix phpcs.

gxleano → made their first commit to this issue’s fork.

Everything looks good to me @kim.pepper.

I've just created a new issue on SQLite provider where it should be also changed after this change.

See: https://www.drupal.org/project/ai_vdb_provider_sqlite/issues/3549485 📌 Change filter typehint from string to mixed Needs review

gxleano → created an issue.

Taking a look on this.

Hey Scott!

I've added some changes and tested the solution, now grouping is working as expected on my side.

Maybe we should just create a new release for 2.x and point this issue to it.

All good for me Scott.

I added some changes to fix phpcs and cspell, the rest is fine.

gxleano → made their first commit to this issue’s fork.

Hey @herved

The jQuery dependency here is quite minimal, but we'd need to refactor the once() usage and update all the jQuery calls.

So, if you have the time, feel free to try it :)

gxleano → made their first commit to this issue’s fork.

Thank you Dieter to keep an eye on this!

It will be included in the next release 1.2.42

gxleano → made their first commit to this issue’s fork.

Thank you @saidatom @dieterholvoet @alorenc for you contribution!

It will be added to the next release 1.2.42

gxleano → made their first commit to this issue’s fork.

Thanks @herved and @pcambra,

It will be included in next release 1.2.42

Testing on AI and AI Agents 1.1.x with Drupal 10.5.2 the changes make it works.

Setup to be reproduce it

- Drupal 10.5.2
- AI 1.1.x (AI Assistant API, AI Search and AI Chatbot)
- AI Agents 1.1.x
- Action (Core module, so it just needs to be enabled)
- Key latest_version
- Search API latest_version
- Ollama Provider (ddev addon) latest_version
- Milvus VDB Provider (ddev addon) latest_version

After applying the MR into AI module the error from the next pages is gone:

/admin/config/ai/explorers/tools_explorer
/admin/config/ai/explorers/chat_generator
/admin/structure/ai_agent/add

Moving the issue to RTBC, thanks everyone involved.

Tested with https://www.drupal.org/project/ai/issues/3543112 📌 Make model id required on the OpenAiBasedProviderClientBase Active applied to AI module and everything works as expected.

The code also look good to me Marcus.

Thanks!

It looks good to me Marcus!

Thanks!