dsh-relay-models

When you wire up a custom API provider in DeepSeek Harness, every model demands manual entry of its context window, output limit, input format, reasoning mode, and compat settings. The pain compounds quickly if you route DeepSeek calls through a relay or a self-hosted endpoint where models appear and change on a weekly cadence. dsh-relay-models was built to remove exactly that repetitive legwork: the moment a model ID is discovered, the plugin cross-references it against the pi.dev catalog and populates all capability fields in a single pass.

The workflow stays intentionally lightweight. After you add a custom provider and let DSH discover the available models, any ID that matches a catalog entry receives its full capability profile automatically. IDs that do not match present a dropdown so you can pick a reference model from the catalog by hand, guaranteeing that every model in your lineup carries a complete and consistent parameter set ready for inference.

If your day-to-day stack already depends on a relay or self-hosted gateway to serve DeepSeek-series models, this plugin lets you stop re-calibrating parameters from scratch each time a new model lands, so you can spend that saved time on prompt engineering and real inference tuning instead.