DeepSeek Harness setup with Qwen 3.8 27B on Ollama

This tutorial sets up DeepSeek Harness (DSH) on Windows with Qwen 3.8 27B running through Ollama. The presenter installs Node.js, launches DSH with npx @deepseek-ai/dsh web, and opens its dashboard on port 3080. The demonstrated DSH release is an RC build described as beta, so the presenter warns that users may encounter issues.

The local LLM connection uses a custom provider with http://localhost:11434/v1 as its base URL. The walkthrough enters a placeholder API key string and a model ID that must match the name in Ollama. Users need to download the model locally first; this setup does not require a DeepSeek API key.

The presenter explains Standard, PTC, Minimal and Creator modes, then chooses Standard with workspace write permissions. The walkthrough also covers slash commands, context compaction and plugins that can be enabled or disabled individually.

For the coding assistant demonstration, the presenter requests a personal portfolio, answers questions about its contents and style, and asks for revisions to remove placeholders. The agent creates an index.html file. The presenter praises the resulting details and links, but the demonstration does not independently verify their accuracy. Deployment to Vercel requires browser login and approval for a CLI command outside the workspace. The presenter reports a working public URL after an earlier suggested URL fails.