v11

What's New

Llamatik Integration

• Replaced custom C++ JNI llama.cpp bridge with Llamatik library (com.llamatik:library:0.16.0)
• Rewrote LlamaCpp and LocalLlmClient to use Llamatik's LlamaBridge API with direct ChatML prompt formatting
• Local model auto-loads from disk on first use if downloaded

Local Model Update

• Switched from Qwen3-4B (~2.5 GB) to Qwen2.5-1.5B-Instruct (~753 MB) for faster downloads and lower memory usage

Tinfoil TEE Fixes

• Fixed base URL (inference.tinfoil.sh instead of api.tinfoil.sh)
• Corrected model IDs: kimi-k2-5, deepseek-r1-0528
• Added gpt-oss-120b as a new Tinfoil model option
• Added error logging for stream failures

Multi-Provider LLM Support

• Added OpenRouter, Tinfoil TEE, and local on-device as selectable LLM providers
• Provider selection in onboarding and settings
• OpenAI format adapter for Tinfoil and local providers

Previous Changes

• ClaWHub threat analyser
• Skill creator and refinement skills
• Improved memory algorithm
• Heartbeat log persistence and in-app viewer
• Dynamic AI name throughout the app
• Insufficient balance dialog with Gas Paymaster deep link
• Wallet signing on Base network
• Premium LLM gateway with wallet signature auth
• Configurable heartbeat interval (5-60 min)
• BeanShell code execution skill
• Termux skill integration
• XMTP messaging support
• Wallet and contacts integration
• And many more improvements since initial release