Trending on GitHub

The SQLiteAI waste repository enables running the Kimi K3 large language model using only 29 GB of RAM. The inference performance is recorded at approximately 0.50 tokens per second. This approach demonstrates significant memory efficiency for LLM deployment without requiring high-end GPU hardware.

• Kimi K3 fits in 29GB RAM, enabling consumer-grade hardware inference
• SQLiteAI waste repo provides the necessary tools for this optimization
• Speed is ~0.5 tok/s, suitable for batch or low-latency tasks
• No GPU required, reducing infrastructure costs for specific workloads

This GitHub repository provides a skill router pack designed to guide AI coding agents through reverse engineering and penetration testing workflows. It features on-demand toolchain bootstrapping and a self-evolving knowledge base to handle targets like APKs, binaries, and frontend JS encryption. The system integrates with clients such as Claude Code, Cursor, and Cline to automate complex security research tasks.

• Directs AI agents through specific reverse engineering and pentesting scenarios.
• Bootstraps required tools on-demand based on the detected target type.
• Maintains an evolving knowledge base to improve future routing accuracy.
• Supports major AI coding clients including Cursor, Claude Code, and Cline.
• Automates handling of APKs, binaries, and encrypted frontend JS challenges.