System requirements

16 GB of RAM is a practical starting point for local speech, AI summaries, and chat on an Apple silicon Mac or a Windows PC with dedicated graphics. An SSD helps models load faster. More memory is useful for larger models and heavy multitasking, but 32 GB is not a requirement for local AI.

Recommendations, not a performance guarantee

These recommendations draw on Kimiko’s model sizes, development measurements, and experience using the app on a 16 GB M5 MacBook Air and 16 GB Windows PCs with dedicated graphics. They are not tested minimums for every computer or model. Your chosen models, meeting length, enabled features, and other open apps all affect performance. Try a representative meeting before relying on a setup for important calls.

Supported computers

Computer	Download compatibility
Mac	Apple silicon, macOS 14.4 or later
Windows	64-bit Intel or AMD PC

For local processing on Windows, we recommend current Windows 11 and a recent Intel Core i5 / Core Ultra 5 or AMD Ryzen 5 class processor or better. On Mac, choose an M-series computer with the memory shown below.

Kimiko’s public downloads are for Apple silicon Macs and Windows x64 PCs. See installation for the installers and permissions for microphone and system audio access.

Choose memory for the work you want to do

RAM is your computer’s working memory. These figures describe the computer’s total installed RAM, including room for the operating system and everyday apps.

Your setup	Recommended RAM
Cloud speech and cloud AI	16 GB for everyday multitasking
Local speech with cloud AI	16 GB
Local speech and AI: Apple silicon Mac	16 GB
Local speech and AI: Windows with dedicated graphics	16 GB
Local speech and AI: Windows with shared graphics memory	16 GB; consider 32 GB for more room

These local recommendations are starting points for smaller models. On Windows, the graphics card and its own memory matter too. Cloud processing lowers local demand, but audio handling, speaker labels, and search can still use memory.

Consider 32 GB or more if you want to run larger models or keep many demanding apps open. This is extra room, not a requirement for using local speech and AI. On an 8 GB computer, start with cloud speech and cloud AI, then test any local features individually. We do not recommend 8 GB for running the full local meeting workflow.

Larger local models can need more than these recommendations. A download marked “3 GB” can use more than 3 GB while running, and Kimiko may need several models at once.

Graphics and storage

A separate graphics card is optional. Apple silicon and compatible Windows graphics can help with local processing. Modern built-in graphics can work too, but their memory comes from the same RAM as your other apps. See memory use on laptops for an explanation of shared memory, also called UMA.

Apple silicon Macs also share memory between processing and graphics. Shared memory alone does not mean a computer needs 32 GB: the hardware, software, and models all make a difference.

Use an SSD and allow at least 10 GB of free space for an initial local setup; 20 GB gives more room for downloads, updates, and trying another model. This is a storage allowance, separate from the RAM recommendations. Larger models, saved recordings, attachments, and additional model downloads need more space. Check the size shown in each model picker.

You need an internet connection for initial downloads, updates, and any cloud providers you choose. Downloaded local speech and AI models can process your content offline.

If your computer struggles

Start with a smaller speech model or local AI model, close demanding apps, and compare a short meeting. You can use local speech with cloud AI, or cloud speech with local AI; the choices are independent.

If you do not need them, turn off Label remote speakers or Live summaries in Settings → Transcription. For dictation, Settings → Dictation → Keep models loaded lets you free memory sooner between uses, at the cost of a short wait when the models next load. The default is 5 minutes.

Cloud processing sends the relevant audio or text to your chosen provider and may have separate charges. It reduces local work, but does not turn off every local feature. Review cloud providers and privacy before switching.