Metka is a macOS app that brings the power of on-device AI to your image library. Drop a folder of images and let Gemma — Google's open-weight vision-language model — analyze each one, generating rich descriptions, semantic tags, and dominant color extraction. All processing happens entirely on your Mac, with zero data leaving your machine.
Whether you're a photographer cataloging thousands of shots, a marketing team organizing product imagery, or a designer building mood boards, Metka adapts its analysis to your specific context through customizable AI prompts and role-based presets.
Powered by Google Gemma 3 running locally via MLX Swift. Choose from Light (4B), Medium (12B), or High (27B) models to match your hardware.
Analyze entire folders of images in one go. Progress tracking, dock badge updates, and haptic feedback keep you informed throughout.
Automatically identifies dominant colors in each image from a curated palette of 15 colors, enabling powerful visual filtering.
AI-generated semantic tags let you search and filter your library by content, not just filenames. Sidebar filtering by tag or color makes discovery effortless.
Define the AI's role and context for tailored analysis. Built-in presets for Marketing, Social Media, and Inventory workflows — or write your own.
All processing happens on-device using open-weight models. Your images never leave your Mac — no cloud APIs, no data collection, no compromises.
Last updated: February 2026
Metka processes all images entirely on your Mac using locally downloaded AI models (Google Gemma 3). No images, descriptions, tags, or any generated metadata are ever sent to external servers or cloud services. Your data stays on your device at all times.
Generated descriptions, tags, and color data are saved as text files alongside your images in a local "Output" folder. Metka does not maintain any separate database and does not collect or store any personal information.
Metka downloads open-weight AI models from Hugging Face (mlx-community). This is a one-time download per model. The models run entirely on your Mac's hardware — no cloud inference is used.
Metka does not include any analytics, crash reporting, or tracking SDKs. The app does not collect usage data of any kind.
The only network activity is the initial model download from Hugging Face. After that, Metka functions completely offline. No advertising SDKs or tracking frameworks are included.
For privacy-related questions, contact us at email.
Need help with Metka? Have a feature request or found a bug? Get in touch.
For support inquiries, bug reports, or feature requests: email
Metka requires Apple Silicon (M1 or later). The Light model (4B) runs well on any M1 Mac with 8 GB RAM. The Medium model (12B) is recommended for M2 Max or newer with 24+ GB RAM. The High model (27B) requires M2 Ultra or M3 Max with 64 GB RAM for optimal performance.
No — never. Metka uses on-device AI models (Google Gemma 3) via Apple's MLX framework. All analysis happens entirely on your Mac. Your images and generated metadata stay local.
Metka supports all image formats recognized by macOS — including JPEG, PNG, HEIF, TIFF, WebP, and RAW formats from major camera manufacturers.
Absolutely. Metka lets you define a custom role and context prompt that shapes how the AI interprets your images. Built-in presets for Marketing, Social Media, and Inventory give you a starting point, or you can write fully custom prompts.