Release Notes

Stay updated with the latest improvements, optimizations, and new features of AI Troll.

New

Version 2.1.0 (Current Stable)

Released: July 2026

A feature release introducing fully offline translation and a reworked theming system, plus a deep fix for Gemma 4 chat formatting.

  • Translator Mode: Turn any multilingual model into a private, fully offline translator on its own dedicated page. Choose a language pair or use auto-detect, swap languages with one tap, and copy the result — translations are never stored and never leave your device.
  • Gemma 4 Chat Template Fix: Prompts are now rendered by the model's own embedded chat template, so Gemma 4 — and future model families — respond correctly instead of repeating greetings.
  • Stability: Fixed a multi-turn cache corruption affecting models with unrecognized chat templates, and control tokens no longer leak as raw text into the chat.
  • Theming Upgrade: A new Day/Night display mode pins the app to light or dark colors independently of the device setting, and the Custom theme now offers separate Day and Night values for all 8 core colors, with automatic night derivation and contrast warnings.
  • Model Management: Models that can translate are marked with a Translator chip and can be filtered with one tap, and the Browse list now verifies each file's real GGUF header — companion files that are not runnable models no longer appear as downloadable.

Version 2.0 (Outdated)

Released: June 2026

Our biggest update yet! Introducing global accessibility and optimization for the next generation of mobile AI models.

  • 12-Language Support: Full interface localization for 12 major languages, including complete RTL (Right-to-Left) layout mirroring for Arabic.
  • Expanded Model Support: Optimized compatibility with Google's Gemma (including Gemma 4), Meta's LLaMA, and Microsoft's Phi models, featuring auto-detected templates and KV Cache preservation.
  • Paginated Model Browsing: HuggingFace model catalog search now retrieves results in smooth, paginated batches to eliminate search lag.

Version 1.0 (Outdated)

Released: January 2026

The initial public release of AI Troll, establishing private, secure, and fully offline AI chat on Android.

  • Native llama.cpp Integration: High-performance local JNI inference engine for running quantized GGUF models on mobile devices.
  • WorkManager Background Downloads: Download large model files in the background with auto-resume and with completion notifications.
  • Local Model Importing: Import any custom GGUF models directly from your device storage.
  • Customization & Safety: In-app theme configuration (8 colors), local content flagging for safety, and clear RAM safety checks.