DeepSeek Harness setup with OmniRoute free API routing

This tutorial connects DeepSeek Harness to OmniRoute, a self-hosted router that sends requests to external model providers. The router runs on the user's computer, but inference uses provider APIs rather than a local LLM. The speaker demonstrates a tower defense game and spreadsheet app, then requests a self-contained HTML Pong game.

Setup covers installing OmniRoute through npm, starting its dashboard and server, and adding a custom provider in DeepSeek Harness with the OpenAI completions protocol. An initial auto model provides a quick connection test. The speaker then changes the dashboard's default password and connects OpenRouter and NVIDIA NIM through API keys, importing only free models.

The combo workflow groups models behind one selectable endpoint and uses round-robin routing. The speaker tests the endpoints, removes failing entries, and fetches the resulting combo in DeepSeek Harness. Passing a test does not guarantee continued availability; the tutorial recommends checking the pool periodically.

Provider allowances vary. The tutorial also warns that prompts go to external services, providers may object to this routing, and response speed and quality vary. The speaker recommends experimentation rather than production use.