Skip to content

Releases: TZYCeng/dsh-model-visibility

1.9.1

Choose a tag to compare

@TZYCeng TZYCeng released this 04 Nov 04:56

Register the client bundle under the scoped package name

  • client.js registers itself as @tzyc/dsh-model-visibility; it still said dsh-model-visibility, so the loader reported loaded without registering "@tzyc/dsh-model-visibility" and the settings page stopped syncing. The host resolves the bundle by package.json name, so the two must match.
  • New test/client-bundle.test.mjs runs client.js against a fake __ModuleLoader__ and asserts the registered id equals package.json name, and that the factory returns { inject, apply }. A future rename that misses the bundle now fails the suite instead of failing in the app.
  • Docs: README version badge, HANDOVER release record, and pitfall 25 — a scoped rename must touch client.js, cordis.patch.yml and the profile row together.
  • Tests: 74 passing.

1.9.0

Choose a tag to compare

@TZYCeng TZYCeng released this 03 Nov 15:30

Manual capability overrides, group collapse and a new README

  • Edit capabilities per model: a new "edit" button on every row sets kind, context window, input modalities and reasoning-effort levels (including a default level, or "no reasoning"). Overrides live in $DSH_HOME/dsh-model-visibility/overrides.json and outrank every automatic source.
  • Write back to the provider: the editor can write input, contextWindow and reasoningEfforts into the llm-pi-ai provider config for that model, so the composer, the adapter and the settings page read the same facts.
  • Collapsible provider groups: collapse one group from its title, or all of them from the toolbar; the state is kept in localStorage and never touches model visibility.
  • Hard capability re-fetch: "re-fetch capabilities" drops the capability cache, the relay probe cache and any in-flight scan, then asks the adapter and every relay again.
  • README rewritten with new screenshots; the page now documents search, filters, overrides, corrections and the image-input alignment in one place.

1.6.0

Choose a tag to compare

@TZYCeng TZYCeng released this 02 Nov 09:41

Search and capability filters

  • The settings page gains a name/ID search box, Multimodal and Reasoning toggles, and a minimum context-window filter (128K / 256K / 512K / 1M), all combinable with the existing kind filters.
  • normalizeModelId now strips stacked request suffixes such as :free:batch one at a time, so catalog lookups stay stable for providers that chain variants.
  • Docs and package metadata for 1.6.0.

1.5.0

Choose a tag to compare

@TZYCeng TZYCeng released this 01 Nov 13:10

Per-model capability tags

  • Tags for reasoning support, every published reasoning-effort level, context window, input modalities and model kind, merged from six sources in priority order: settings document, adapter, the relay GET {baseURL}/models response, the built-in pi-ai / models.dev catalog, model-name inference, and vendor corrections.
  • Solid borders mean a declared value, dashed mean inferred; hovering a tag names the source. Generation, embedding, rerank, audio, classifier and translation models only show the facts that apply to them.
  • Opt-in image-input alignment: relay routes whose models declare text-only input accept images without editing any configuration file.
  • Hidden list persisted in the profile patch and mirrored to $DSH_HOME/dsh-model-visibility/hidden.json.

Also: package metadata, screenshots.json and README screenshots for the plugin marketplace.