1.8.0

What's New

Features

• Vision — send a photo and chat about it. Attach an image from your gallery or snap one with the camera, then ask your model what it sees. Works with Gemma, Qwen, and Ministral vision models; the image module downloads automatically with the model.
• Tabbed generation settings — generation parameters, system prompts, and tools now live in one tabbed chat bottom sheet.

Models

• Gemma 4 (E2B / E4B) ships with vision enabled on the default QAT Q4_0 build.

Under the Hood

• Vision runs fully on-device via llama.cpp's mtmd library.
• Enabled the Vulkan GPU backend; updated llama.cpp to b9621.
• Device-aware vision encoder placement (CPU on PowerVR GPUs for correct, faster output).
• Localized the app, store listing, changelog, and screenshots into 27 languages.

Bug Fixes

• Fixed a UTF-8 crash in the JNI generation callback.
• Load a picked image from a local copy instead of the picker URI.