“Local Code” by Draw Things is now open-source. It comes with a built-in Python environment, allowing you to edit spreadsheets, create presentations, and generate images—all entirely on your own machine. The starter package features the Qwen 3.8 27B model, which is approximately 10GB in size.
A quick note: Using a local model ensures your data never leaves your device. Will you give it a try?
Running a 27B-parameter model on 16GB of RAM hinges on the actual memory footprint after quantization and the requirements for context length. With the model itself occupying roughly 10GB, there is limited headroom left for long-context windows once system and application memory usage are accounted for. For tasks like processing spreadsheets, slide decks, or generating images, performance depends on the success rate of tool calls and the permissions granted to the local Python sandbox. Ensuring that data never leaves the device requires that plugins, fonts, and templates also do not transmit information externally. Ultimately, speed and accuracy must be evaluated using a representative set of real-world files; simply confirming that the model can fit into memory is not enough.