This app succesfully creates a session and keeps context, the the built-in AI will remember what you asked.

This is the first program to make the local Gemini instance fully remember the context, as far as I am concerned.

This app does not uses any wrappers for the built-in AI.

You can visit a live version here, but keep in mind that you'll need to have a proper environment (as described below) to use it.

This AI accesses an offline API provided locally by Chrome versions 128 or above.

1. Ensure Compatibility:

   • Verify you are using Chrome version 128 or above. If not, consider downloading:
     • Canary (nightly builds)
     • Dev (for developers)

2. Enable Required Flags:

   • Open Chrome and type chrome://flags in the URL bar.
   • Set the following flags:
     • Enables optimization guide on device to Enabled BypassPerfRequirements
     • Prompt API for Gemini Nano to Enabled

3. Download the Optimization Guide:

   • Relaunch Chrome.
   • Open Chrome again and type chrome://components in the URL bar.
   • Find the component named Optimization Guide On Device Model and download it (approximately 1.5 GB).

4. Final Steps:

   • Relaunch Chrome.
   • If everything is set up correctly, you can start using the AI. Otherwise, window.ai will be undefined.