The APEX Working Copy feature is only focused on applications within the workspace schemas. I'm not sure how it could work as you ask, but perhaps make the idea submission for further discussion : apex.oracle.com/ideas

Yes

Yes, you can import vector data like any other built-in Oracle data type using the familiar utilities like data pump or other tools you're familiar with from Oracle.

At the moment the Select Many doesn't provide that capability, if that is something you would like to see you could add an idea on the ideas app : apex.oracle.com/ideas

today, the QR item type is only for generating QR code imagery

Multi Modal ONNX models do that. It’s in roadmap for the Database 23ai to support multi modal embedding models

You can lock a page in main using the familiar lock icon in the page designer. Other users will see that the page is locked in page if they click on the warning icon in page design when there are visiting that page in a working copy.

See the earlier answer about the 23ai availability. In short - this calendar year.

Grouped items, but not multi-level trees.

No, does not require Oracle Text index.

It can support any language. depends on the ONNX model you load to convert the text to vectors.. As long as the ONNX model can convert the text to vectors, once we have vectors, language becomes irrelevant.

You can create indexes as needed to improve the speed of the query. Using the dev toolbar Debug mode, you can enable Info level debug and see what queries are performed for the search and then ensure that column has an index on it in your underlying table.

This is a complex question and probably deserves its own Office Hours session. Perhaps you can post it in https://apex.oracle.com/forum to start the discussion. It seems you would want to switch APEX authentication to Fusion IDCS.

This is an excellent question to post on https://apex.oracle.com/forum

The user who wants to use dbms_vector to create credential needs to have that privilege. In case of cloud, the Admin user does not have that privilege. That’s why that error. A fix is coming in 23.5 where the cloud user will be able to create credentials using dbms_vector. The workarounds are:

1. if you have access to sys, you can grant that privilege to the user who wants to create credentials. Now I am not sure whether you have access to sys in DBCS or ADBS or ADWS.
2. you can use database provider (aka onnx model) instead of using 3rd party providers. In that case, you don’t even need credentials.

Please post this question on https://apex.oracle.com/forum

The latest version I can find info about on the web appears to be 1.16.1 at https://onnx.ai/ In the "News" section of this website, I'm not seeing mentions of a v2.

Not sure if this helps, but one reply concluded with " I believe the only work-around to leverage the OCI Gen AI Service is to leverage a 23ai DBCS instance";also, from Connor - "Sounds like a bug in that particular instance. My (autonomous) 23ai is now at 23.5 and no additional privs were needed for dbms_vector, eg"

The DBMS_VECTOR_CHAIN package provides utilities to do chunking inside the database.

APEX Developers currently have 2 choices for RTE page items: TinyMCE and Oracle Rich Text Library (based on CKeditor) - I am unaware of any specific improvements that have been made with regards to "huge content" LOB support - this sounds like you should consider implementations with dynamic content regions perhaps. I'd recommend you post a question on our forum for further discussion with the community apex.oracle.com/forum

You can still create Vectors outside the Database and use commercial and open-source models with either REST Calls or Local Libraries. Please refer to the FAQ Page for 23ai for more details. https://www.oracle.com/in/database/faq/

The Select List uses the browser's native select controls, where the Select One provides a flexibly-styled, autoreduce-as-you-type functionality. So you have the option to use either depending on your requirements.

If they can be done in Database, then they can be done in APEX - APEX is just the UI/veneer on top of the Database from which it is deployed.

The Combobox allows manually entered items. The Select Many does not have manually entered items. Both support multiple values.

You can configure a maximum number of entries, but there is no hardcoded limit by default. If your list has many, many entries, then you can force the user to type at least N characters first and do search as you type so that it's never retrieving the entire list.

Think of the ONNX file as the "bytecode" that describes the algorithm that converts input text into a vector embedding. It is a kind of "black box" computing machine that has been trained on a dataset and its calculations expressed in this industry standard format. You can read more about the ONNX standard at https://onnx.ai/ but typically a user of an ONNX model wouldn't need to understand the ONNX standard itself. This is similar to how you can run a Java program using the compiled Java bytecodes without really understanding anything about what the bytecodes represent. It's just a standard, portable way to represent a "vector calculation program" that encodes semantics about input text based on a particular training model.

Currently Database 23ai is only available in Always Free databases in 4 regions (Ashburn, Phoenix, Frankfurt, London). It is planned to become available in Paid Tier databases in all regions this calendar year. In the mean time, you can also use the Autonomous Database 23ai Free Container image - see https://blogs.oracle.com/datawarehousing/post/autonomous-database-free-container-image

I loop through my main table and store the rach column information as chunk (different records)

To the best of my knowledge, the determinism of embedding content into vectors using an ONNX model depends on several factors, but generally the same input should generate the same output vector.

Cache contents are automatically loaded when accessed and refreshed when the data changes. Oracle True Cache supports all Oracle Database APIs and data types, including relational, JSON, text, spatial, graph, and vector data types