Feature Requests

Kerling Windows does not behave like a typical application window. It cannot be resized or positioned on one side of the screen or set to full screen to read responses properly.

The On-Device Foundation Model support released in 2.5.0 is already great! However, adding the Private Cloud Compute Server Foundation Model would complement that perfectly, because: It’s probably the most secure model for sensitive data right after offline models It’s much more capable (at eye level with 70B models, instead of only 7B) It’s entirely free for users https://machinelearning.apple.com/research/introducing-apple-foundation-models If it’s possible to integrate it, that would be really cool. Thank you!

Selecting one or more local text files (like .md) within the Presets configuration would make that feature much more versatile. For example, this would allow me to: Collect my previous emails using Obsidian and use that note within Kerlig to ensure the model answers with my tone of voice. Reference documentation help files of my app or product to quickly answer customer questions. I could also copy and paste the content of these documents there, but then they would become stale quickly, and I would have to update them manually on a regular basis. It would generally help give the AI model much more context. This would be really cool! Thank you!

Hi Kerlig team, It would be best if google AI mode is supported in Kerlig, I know gemini is supported but, Google AI mode is more helpful in search and upto date contents.

Make Kerlig recognize specific code block tags and treat their contents as files. Example: A custom prompt that turns selected text into an ICS calendar event would output: ```ics EVENT DATA GENERATED BY PROMPT ``` If the code block is tagged ics , Kerlig extracts its contents, generates an .ics file, and shows the file inline instead of the code block. Clicking opens it in the default macOS app; right‑click provides options to save, open with another app, or reveal the raw contents. Similar to text correction feature, TAB key could be used to switch between file view and raw code block view.

Just like the integration with apple mail I would like other apps like Spark Desktop but preferabel a generic way to use apps as context.

This is a fantastic model but the problem is that the returned output currently includes the thinking block, which is not ideal when doing text transformations. I still want the model to think, I just don't want the output to include the thinking block. The easiest way I can think of doing this (if the model itself can't be told to not include the thinking block in the output) is to have Kerlig itself have the option to filter for <think>...</think> and hide that from the output. This is specific to Qwen3, I know other thinking models annoyingly use a different syntax for tagging the thinking block.

add functionality that will return an image as a response if the generate image promt is used

I'd like to be able to block/add a warning certain models based on the macOS focus mode. E.g. when in work I want to avoid certain models or on the reverse not accidentally use work models for personal requests. ideally this should both work as a filter and/or just a visual warning if e..g a work model is selected when in a personal focus mode

I know this is a big lift but it would be awesome if somehow Kerlig was my main MCP client with an option to choose the inference server like ollama or OpenAI