Guide - How to Use Gemini AI Assistant on GrapheneOS

Ashoka74

There is a parameter related to AICore activate Gemini Nano from developer mode:
https://beebom.com/how-enable-gemini-nano-pixel-8-8a/

Would it be possible to add it to Grapheme for the next release on PixelPro8? Offline transcription is not possible on my end even with Gemini and GooglePlay installed!

Thanks for the hard work! If there is a workaround I am listening full ears

de0u

Ashoka74 I suspect it's not possible to change an AICore developer option on GrapheneOS because, I suspect, AICore isn't part of AOSP (it appears to be part of the "Google AI Edge SDK").

Amk

Would be great if someone from the development team could comment on this. I'm really curious about using on-device generative AI, which could actually lead to more private assistants compared to the one made by Google.

Roger

Amk 15k USD gets you hardware to train your own:

https://tinygrad.org/#tinygrad

Ashoka74

https://github.com/google-gemini/generative-ai-android

https://ai.google.dev/edge/mediapipe/solutions/setup_android?hl=fr

These links might be helpful !

Anais

You want the grapheneos team to add Gemini when they've just removed vanadium's Google translator?

I'm French, I must have misunderstood, but you want Gemini in grapheneos?

Dumdum

Anais You want the grapheneos team to add Gemini when they've just removed vanadium's Google translator?

I don't think they said anything that even remotely suggested this.

you want Gemini in grapheneos?

Quite a difference between on and in GrapheneOS. People are free to decide if they want Gemini on (their) GrapheneOS.

Personally, the day they integrate this, I'm going to quit GrapheneOS right away.

There's good news then. You won't have to quit as they won't integrate it, for the reasons being detailed here and here.

Ashoka74

Anais With Pixel 8 and 8 Pro, you can run Gemini Nano locally (no internet access).
The idea I guess would be to take advantage of the phone's GPU to run small open-source models locally
(Llama3, Gemma, Mistral..)

Anais

Personally, the day they integrate this, I'm going to quit GrapheneOS right away.

Ashoka74

Update:
I managed to get a Gemma-2b model running locally following this tutorial:
https://github.com/google-ai-edge/mediapipe-samples/tree/main/examples/llm_inference/android

The speed is quite good and I am running it in plane mode so no data leaks for sure!

de0u

Ashoka74 Thanks for reporting back! It's great for users who want to experiment with LLMs on Pixels running GrapheneOS to have a clear starting point.

L8437

@Tryptamine hi mate, are your original instructions still what you would do?

Is there anything different you would do?
Such as different permissions/settings?
If so, why?

Cheers

Elsewhere3987

L8437 I followed the instructions but never got it to work.
https://discuss.grapheneos.org/d/15405-google-gemini-app-error

spammerofspam

I tried this again recently, and Gemini has started working for me. I followed the same steps as previously, but it didn't work previously. Guess it was a recent change in the app?