Draw Things Launches Public Beta of Local Code Execution Tool

Draw Things has launched the public beta of Local Code, which integrates local LLM inference with various development tools while complying with macOS App Store sandbox requirements.

Performance and Features

• Inference Speed: Based on the Draw Things inference engine, prefill speeds for supported models have improved by 1.2x to 1.6x. On the M5 Max chip, it achieves approximately 980 tokens per second for the Qwen 3.8 27B model and approximately 900 tokens per second for the DeepSeek 4 Flash 0731 model.
• Sandbox Environment: Operates within a secure, independent package via Apple's App Sandbox and Hardened Runtime.
• Integrated Tools: Various development tools such as Python, Node.js, Lean, and Bash are built-in by default, allowing immediate use without separate configuration.

Usability

• Download the app via TestFlight and install models to use it immediately without sign-up or login. This contrasts with tools like OpenCode, which require login for web search tools. It also supports multitasking, allowing multiple tasks to run simultaneously on a single M5 Max laptop.