I've spent several hours trying to get the Oracle_OS app to recognize a custom Gemma 3 model (.gguf format) and a personality file (Oracle_OS.md). Despite placing the files in all the locations specified by the documentation and the app's own error message, the app consistently fails to find the model. This report details my environment, the steps I took, and the troubleshooting I attempted.
Environment:
Device: Realme P3 Ultra
Android Version: 15 (Realme UI 7.0)
ASI Version Tested:
Oracle_OS.apk from v2.1.2 release (~97MB)
app-debug.apk from the "HUD patch" release (~900MB)
Model File: gemma-3-4b-it-abliterated-q4_k_m.gguf (2.49 GB)
Personality File: Oracle_OS.md (4.2 KB), created manually with a custom system prompt.
Steps to Reproduce:
Downloaded the app-debug.apk (HUD patch) and later the Oracle_OS.apk from the v2.1.2 release and installed them.
Downloaded the gemma-3-4b-it-abliterated-q4_k_m.gguf model file directly to the phone's Internal Storage/Download/ folder.
Created an Oracle_OS.md file with a custom system prompt and placed it in Internal Storage/Download/.
Launched the app.
Observed Behavior:
A notification appears with the error: ERROR: No model found. Place .literltlm or .gguf in app folder/Downloads.
The app's main interface continues to show "Gemma" but is non-functional.
Troubleshooting Attempts:
Following the error message and documentation, I attempted to place the files in various locations. All of the following attempts failed with the same error.
Placing files in the root Download folder:
Ensured both Oracle_OS.md and gemma-3-4b-it-abliterated-q4_k_m.gguf were in Internal Storage/Download/.
Force-stopped and relaunched the app.
Placing files in the app's private data directory (using ADB):
Used adb shell and run-as com.gemma.api to copy the files.
Created the files/ and files/models/ directories.
Copied Oracle_OS.md to /data/data/com.gemma.api/files/Oracle_OS.md.
Copied the .gguf model to /data/data/com.gemma.api/files/models/gemma-3-4b-it-abliterated-q4_k_m.gguf.
Verified file placement with ls -la. Files were confirmed to exist with correct permissions.
Force-stopped and relaunched the app.
Placing files using a file manager with Shizuku (MiXplorer):
Granted MiXplorer root-like permissions via Shizuku.
Navigated to /Internal Storage/Android/data/com.gemma.api/files/.
Observed that the directory was completely empty. This was after multiple ADB copy attempts that reported success, indicating a possible permission or file system barrier.
Expected Behavior:
The Oracle_OS app should automatically detect the .gguf model file placed in either the root Download folder or its own private files/models/ directory. Upon successful detection, the model should load and the app should become functional, with the AI adopting the personality defined in Oracle_OS.md.
Additional Notes:
The same behavior was observed on both the v2.1.2 release and the larger app-debug.apk.
I have a significant amount of experience with local LLM deployment (Ollama, Termux), even tried all the different Claws available in github. b
and have successfully disabled Realme's aggressive process killers to ensure a stable environment. I do not believe this is an environmental issue.
Thank you for your time and for building this project. Please let me know if I did something wrong or is there something I need to know why it didn't worked.
I've spent several hours trying to get the Oracle_OS app to recognize a custom Gemma 3 model (.gguf format) and a personality file (Oracle_OS.md). Despite placing the files in all the locations specified by the documentation and the app's own error message, the app consistently fails to find the model. This report details my environment, the steps I took, and the troubleshooting I attempted.
Environment:
Device: Realme P3 Ultra
Android Version: 15 (Realme UI 7.0)
ASI Version Tested:
Oracle_OS.apk from v2.1.2 release (~97MB)
app-debug.apk from the "HUD patch" release (~900MB)
Model File: gemma-3-4b-it-abliterated-q4_k_m.gguf (2.49 GB)
Personality File: Oracle_OS.md (4.2 KB), created manually with a custom system prompt.
Steps to Reproduce:
Downloaded the app-debug.apk (HUD patch) and later the Oracle_OS.apk from the v2.1.2 release and installed them.
Downloaded the gemma-3-4b-it-abliterated-q4_k_m.gguf model file directly to the phone's Internal Storage/Download/ folder.
Created an Oracle_OS.md file with a custom system prompt and placed it in Internal Storage/Download/.
Launched the app.
Observed Behavior:
A notification appears with the error: ERROR: No model found. Place .literltlm or .gguf in app folder/Downloads.
The app's main interface continues to show "Gemma" but is non-functional.
Troubleshooting Attempts:
Following the error message and documentation, I attempted to place the files in various locations. All of the following attempts failed with the same error.
Placing files in the root Download folder:
Ensured both Oracle_OS.md and gemma-3-4b-it-abliterated-q4_k_m.gguf were in Internal Storage/Download/.
Force-stopped and relaunched the app.
Placing files in the app's private data directory (using ADB):
Used adb shell and run-as com.gemma.api to copy the files.
Created the files/ and files/models/ directories.
Copied Oracle_OS.md to /data/data/com.gemma.api/files/Oracle_OS.md.
Copied the .gguf model to /data/data/com.gemma.api/files/models/gemma-3-4b-it-abliterated-q4_k_m.gguf.
Verified file placement with ls -la. Files were confirmed to exist with correct permissions.
Force-stopped and relaunched the app.
Placing files using a file manager with Shizuku (MiXplorer):
Granted MiXplorer root-like permissions via Shizuku.
Navigated to /Internal Storage/Android/data/com.gemma.api/files/.
Observed that the directory was completely empty. This was after multiple ADB copy attempts that reported success, indicating a possible permission or file system barrier.
Expected Behavior:
The Oracle_OS app should automatically detect the .gguf model file placed in either the root Download folder or its own private files/models/ directory. Upon successful detection, the model should load and the app should become functional, with the AI adopting the personality defined in Oracle_OS.md.
Additional Notes:
The same behavior was observed on both the v2.1.2 release and the larger app-debug.apk.
I have a significant amount of experience with local LLM deployment (Ollama, Termux), even tried all the different Claws available in github. b
and have successfully disabled Realme's aggressive process killers to ensure a stable environment. I do not believe this is an environmental issue.
Thank you for your time and for building this project. Please let me know if I did something wrong or is there something I need to know why it didn't worked.