[Other] Why RAG-proceeding feature of the attached files on Android version is removed without any other replacements?

[LoaderTime]

I am currently using Android version of Chatbox, which is v1.16.3.

Before the UI update (versions before v1.16.1, I tested it on v1.15.4) on Android version, once I attached some txt files to the message and send it, it will show it's currently "proceeding".
And those files only used around 16k-17k tokens.

After the update, I reproduced the same attaching operation. But there is no more special message, just a tick like Windows version.
Then, it will use around 60k tokens, which is a large number. And it's the actual number if I just send those files as raw texts.

I tried to use the "knowledge base" feature on Windows version and tested it with those txt files again, using the bge-m3 embedding model from SiliconFlow Api.
Yes, like the things before update, it's around 16k-17k tokens.

I checked out the update logs and issues, and there are no logs or issues for this problem.
There are also no buttons or settings that clearly linked to "knowledge base" or something else in current Android version.
Just an "upload files" button.
I am wondering why this is happening.
Is it just a bug?
Or just something I have missed or mistaken?

[LoaderTime]

I extracted the data for a message with file attachments from the v1.15.4 local SQLite database. The structure was as follows:

json
{
  "timestamp": 1759335207844,
  "files": [
    { "id": "...", "name": "aftermath.txt", "storageKey": "..." },
    { "id": "...", "name": "character.txt", "storageKey": "..." }
  ],
  "tokenCount": 18,
  "id": "...",
  "role": "user"
}

This proves a dedicated processing pipeline existed. The files were handled as structured metadata ("files": [...]), and the final prompt's token count ("tokenCount": 18) was independent of the file size.

Further testing on a conversation migrated from v1.15.4 to v1.16.3 reveals the following:

1. When I continue the conversation with the pre-existing files (attached in v1.15.4), the app correctly uses the RAG process, consuming only 16~17k tokens.
2. However, if I upload a new file in the same conversation using the v1.16.3 UI, it triggers the inefficient process, consuming ~60k tokens.

Speculation:
This proves that the original RAG processing logic was not removed in the update.
It still exists and works for legacy data.
The issue is that the new file upload UI component introduced in v1.16.1 is incorrectly wired to a new, inefficient "raw text" processing path, orphaning the perfectly functional legacy RAG mode.

I also checked out the related directory /proc/<uid>/fd/ and /user/0/data/<package_name>/, not seeing any files related to this.
I've also noticed that the app's data storage has migrated from an accessible SQLite database to WebView-based storage (likely IndexedDB or Local Storage), as indicated by numerous links in the /proc directory.
This change in data architecture likely coincided with the introduction of this issue.

Thank you for your work on this excellent application.
I hope this detailed analysis helps in resolving the issue quickly.