swift-scribe

Thanks for posting this. I tried using your Guidelines instructions format[1] in my app but no matter what, the LanguageModelSession always returns very lengthy / verbose responses. I set the maximumResponseTokens[2] option to various small numbers but it doesn't appear to have any effect. Did you by chance encounter this issue as well? [1] https://github.com/slipboxai/swift-scribe/blob/7952c9e845cff... [2] https://developer.apple.com/documentation/foundationmodels/g...

Despite the Liquid Glass controversy, our team is quite excited about the new on-device ASR model and Foundation Model Framework. It makes building local-first AI apps much easier, it was very easy to build a basic scribe app compared to our first iteration of our meeting note taker. Everything runs locally with zero cloud dependencies. With the option to combine with cloud models for more complex tasks, we hope to see more local-first AI apps. Microsoft just announced Foundry local as well, and it'll be interesting to see where they go with it. There are still some limitations with the foundation model, such as a 4,096 context window, but it will probably get better over time: https://gist.github.com/samhenrigold/3aad01b762ccc87e34e6115...