dsh-awesome-model-setting

Configuring model capabilities in DeepSeek Harness—input modalities, context window, output token cap, and image request budgets—used to mean hand-editing settings.yaml. A single typo in a field name or an extra indentation level and a model silently falls back to defaults, leaving you to hunt for the mismatch.

dsh-awesome-model-setting adds a Model Capabilities page inside the settings panel. Every model under every registered route is laid out in a table with the four key fields as graphical controls. The page shows effective values merged with the adapter's built-in catalog, and labels the source of each value as user-declared, provider-catalog, or route default. You can toggle modalities, type in numbers, or clear a field to restore the default; reset a single model's route defaults, reset an entire group, or pull newly discovered models in via the provider-load button. All writes go through DSH's official settings channel with namespace revision for conflict detection, and changes take effect immediately without a restart.

Built for DSH users juggling multiple provider routes—DeepSeek official, Opencode-Go, self-hosted gateways, or local LM Studio—and especially for anyone who wants to verify at a glance whether a model truly accepts image input or which context window is actually in force, without grepping through YAML line by line.