dsh-model-info-fill

Managing metadata across dozens of custom-provider models—context window, max output tokens, reasoning effort tiers, image input support—quickly becomes a repetitive chore that is easy to get wrong. dsh-model-info-fill removes the manual step: it resolves each model by name against the models.dev capability catalogue and fills in whatever fields are missing. Fields you have already set are never touched, and models the catalogue does not cover fall back to user-configured defaults, so nothing is left blank or guessed at.

Name matching is case-insensitive with a three-level fallback (exact id, provider/ prefix stripped, id-suffix reverse lookup). Auto-fill can fire on every provider save or be triggered as a one-shot pass over all custom models. Unmatched entries are surfaced in a panel where you can adjust the global defaults or fine-tune a single model, including its default reasoning tier for new conversations under that provider. Leaving every reasoning tier unchecked simply marks the model as non-reasoning.

The plugin writes exclusively to the model array under llm-pi-ai.providers.*.models[] and leaves official DeepSeek routing, compat flags, headers, API endpoints, and base URLs completely untouched. A local models.dev cache refreshes silently after seven days; on fetch failure it falls back to the stale cache or the configured defaults without ever blocking a save. It is built for DSH users who maintain multiple custom provider models and want metadata upkeep to be set-and-forget yet still respect manual overrides every time.