SafeCircle AI Playground v1.0.1

v1.0.1 — fix Horizon model token limit

SafeCircle AI Playground v1.0.0

Added

• Initial release of SafeCircle AI Playground for Android
• Horizon Mobile INT4 — fastest on-device safety detection model (Gemma 3 1B, 4-bit, <600 MB)
• Horizon Mobile INT8 — highest-accuracy mobile safety detection model (Gemma 3 1B, 8-bit)
• Horizon Edge E2B — general-purpose agent model fine-tuned from Gemma 4 E2B with 8 GB+ RAM support
• Safety Detection task with real-time conversation risk analysis (grooming, bullying, exploitation)
• LLM Chat task for general conversation with Horizon Edge models
• LLM Prompt Lab for single-turn prompting and experimentation
• Agent Chat — on-device agentic task execution with Skills and MCP tool support
  • 12 built-in skills: calculate-hash, create-calendar-event, interactive-map, kitchen-adventure, learn-something-new, mood-tracker, qr-code, query-wikipedia, read-calendar-events, schedule-notification, send-email, text-spinner
  • MCP server management (add/remove/toggle servers and tools)
  • Skill management UI (enable/disable individual skills)
  • Android intents: email, SMS, calendar events, notifications
  • JavaScript skill execution via embedded WebView
• GPU/CPU/NPU accelerator selection with automatic CPU fallback
• Speculative decoding support for compatible models
• Model import from local storage or URL
• HuggingFace OAuth authentication for gated model downloads
• Benchmark task for measuring model performance
• SafeCircle branding with Horizon AI persona
• Privacy-first: all inference runs entirely on-device, no data sent to servers