deepseek-harness-setup

DeepSeek Harness Setup is a repository that installs a complete DeepSeek Harness stand on a Windows PC. It includes a local model on the user's own GPU, a coding agent that reads and edits project files, a side panel for rendering results such as HTML pages, PDFs, and spreadsheets, and a mobile interface usable from a phone. The repository states that everything runs locally and nothing is sent to a cloud.

The setup is described as a turnkey install: the user copies config.json, runs install.cmd as administrator, and ends up with a working system whose parts already fit together. It also sets up plugins, applies patch layers, tunes the model server for the GPU, and installs shortcuts, an update button, and a full uninstall. The installer picks a model that fits the available VRAM, downloads it, computes the context window, and verifies the result by measurement.

Requirements include Windows 10/11, an NVIDIA GPU with 8 GB VRAM minimum and comfortable from 12 GB, 40 GB free disk, and 16 GB RAM. The configuration is tuned for an RTX 5080 with 16 GB running Qwen3.8 27B at UD-Q3_K_XL. Without an NVIDIA GPU, the stand still installs with a CPU build of the engine, but the model runs tens of times slower.

Daily use includes launching Harness AI from the Start menu to bring up the model and interface, with the interface at http://127.0.0.1:3080. From a phone, access is possible via home network QR code, temporary tunnel, or permanent domain, and the site advises setting an access password before exposing anything. Stopping the stand uses the power button in the top-left corner of the interface, with four levels.