Skip to main content

Built-in AI

Feature Request

Built-in AI would allow local model usage without relying on external providers, which would increase privacy, independence, and the predictability of the app’s behavior. A good direction would be the option to download a lightweight model, e.g. the smallest version of Gemma 4 (E2B or E4B), along with a simple choice between local and cloud modes. It would also be worth noting that this feature should be optional, so the user can decide for themselves about resource usage and where the model is stored.

Posted by Miersetnik ·
4 comments

Log in to comment and vote

Comments4

  • Vowen

    Team•

    Jul 29

    We have experimented on multiple occasions with E2B and E4B and we noticed that it’s not able to follow the formatting instructions consistently to point where it was clear to us it might be frustrating for the user if this was offered in-app. And command mode completely missed on more than 8/10 cases in our tests.

    This is why we have the option to configure “Custom API server”. You can spin up a model locally and point it through that. It’s less ideal, but the current Gemma 4 hallucinates more frequently than can be expected.

    For macOS, we are working on a system to connect with Apple Intelligence directly, and we’re seeing good results with this so far.