v0.3.0 — Audit fixes + type filter chip

Added

• In-menu type filter chip — when the catalog contains 2+ model types and the type prop is unset, a "Filter" chip appears next to the sort button. Click to narrow the list to one type. New props: typeFilter, onTypeFilterChange, showTypeFilter.
• Strict coercion helpers — toBool, toStr, toStrArray alongside toNum. Used internally to replace unsafe as casts throughout the normalizers; exported from both "open-model-selector" and "open-model-selector/utils".
• Venice supportsMultipleImages — added to TextCapabilities; extracted from model_spec.capabilities.
• Lint + format + pre-commit tooling — ESLint 9 (flat config, typescript-eslint + react-hooks + jsx-a11y), Prettier, Husky + lint-staged. npm run lint is a new CI gate.

Fixed

• OpenRouter cache pricing no longer silently dropped — normalizeTextModel now reads OpenRouter's input_cache_read and input_cache_write fields in addition to cache_input/cache_write.
• Capability fields reject non-boolean shapes — an API response like { supportsVision: { enabled: true } } now resolves to undefined instead of being cast to true.
• Sort button ARIA — aria-label describes the current sort state (not the next action); aria-pressed reflects whether "newest" is active.
• Pill contrast — capability pills (vision/reasoning/code/etc.) now meet WCAG AA in both light and dark mode.
• Dark-mode tooltip portal — .oms-hover-content gets explicit dark-mode overrides (background/border/shadow) in both media-query and class-based dark mode.
• Reduced-motion spinner — the loading spinner pulses instead of rotating when prefers-reduced-motion: reduce.

Changed

• BREAKING: TextModel.context_length is now number | undefined — previously the field defaulted to 0 when no context-length source resolved, which was indistinguishable from a model with genuinely zero context. Now undefined signals "unknown". Consumers doing if (model.context_length > 0) are unaffected; anything branching on === 0 will need to handle undefined.
• Icons in icons.tsx default to aria-hidden="true" (overridable via props).
• <ModelSelector> warns in dev mode when rendered with neither models nor baseUrl.
• now reference for deprecation filtering is captured once at mount (was refreshed on every popover open — caused memo cascades).
• handleModelSelect reads from a live ref so its identity is stable across favorite toggles. ModelItem's custom memo comparator works again.
• tsup no longer emits a no-op "use client" banner on the CJS build (ESM still gets it; esbuild places the CJS banner after the runtime preamble where it has no effect).
• CI runs npm run test:coverage — thresholds (70% statements/functions/lines, 65% branches) now actually gate PRs. Baseline: 95.79 / 90.36 / 96.09 / 98.3.

v0.2.0

[0.2.0] — 2026-02-20

Added

• Multi-provider type resolution — defaultModelNormalizer now uses a 6-tier classification strategy to correctly identify model types across providers that use non-standard vocabulary:
  1. Direct match against canonical types (Venice)
  2. Non-text alias mapping (Together AI: "audio"→"tts", "transcribe"→"asr")
  3. Architecture-based inference from output_modalities (OpenRouter)
  4. Heuristic inference from model ID patterns
  5. Text-aliased type (Together AI: "chat"→"text", Mistral: "base"→"text", Vercel: "language"→"text")
  6. Default to "text"
• TYPE_ALIASES export — maps provider-specific type strings (chat, language, base, moderation, rerank, audio, transcribe) to canonical ModelType values; exported from both "open-model-selector" and "open-model-selector/utils" so consumers can inspect or extend
• Extended MODEL_ID_TYPE_PATTERNS — new heuristic patterns: embedqa, imagen, gpt-image, image-preview, kling
• Provider compatibility test suite (provider-compat.test.ts) — 418 lines covering type resolution for Together AI, Vercel AI Gateway, Mistral AI, OpenRouter, OpenAI, Nvidia NIM, Helicone, and Venice AI response shapes, plus defaultResponseExtractor wrapper shapes
• Live provider Storybook stories (provider-live.stories.tsx) — interactive stories for testing against 12 real provider APIs: OpenAI, OpenRouter, Together, Groq, Cerebras, Nvidia, Mistral, DeepSeek, SambaNova, Venice, Helicone, and Vercel
• .env.example — environment variable template for 13 API provider keys used by Storybook live stories and the snapshot capture script
• Provider snapshot capture script (scripts/capture-provider-snapshots.cjs) — fetches and stores real provider API responses for test fixture generation
• .env parser utility (scripts/parse-env.cjs) — shared .env file parser used by both Storybook config and the snapshot capture script
• Storybook provider proxy support — .storybook/main.ts now loads .env keys and configures Vite proxies for cross-origin provider API access during development

v0.1.0

[0.1.0] — 2026-02-13

Initial release of open-model-selector.

Added

• <ModelSelector> component — accessible combobox for selecting AI models, built with cmdk and Radix Popover
  • Managed mode: fetches models from any OpenAI-compatible /v1/models endpoint
  • Controlled mode: accepts a static models array
  • Fuzzy search by name, provider, ID, and description
  • Favorites with localStorage persistence (uncontrolled) or external state (controlled via onToggleFavorite)
  • Sorting by name (A-Z) or newest, controllable or uncontrolled
  • "Use System Default" sentinel option (SYSTEM_DEFAULT_VALUE)
  • Popover placement (side prop: top, bottom, left, right)
  • Custom placeholder text
  • className and ref forwarding
  • Configurable storageKey for localStorage namespace isolation
  • showSystemDefault toggle
  • onChange callback receives the model ID and the full model object (AnyModel | null) — no secondary lookup needed
• useModels hook — fetches and normalizes models from an OpenAI-compatible API
  • baseUrl and apiKey props for endpoint configuration
  • Custom fetcher prop for SSR, proxies, and testing (ref-stored — no memoization needed)
  • Custom responseExtractor for non-standard response shapes
  • Custom normalizer for non-standard model shapes
  • AbortController cleanup on unmount
  • Re-fetches on baseUrl or apiKey change
• defaultModelNormalizer — handles OpenAI, Venice.ai, and OpenRouter response shapes
  • Extracts provider from slash-separated IDs or owned_by
  • Resolves context_length, context_window, and Venice model_spec.availableContextTokens
  • Normalizes pricing from pricing, metadata.pricing, cost, and Venice model_spec.pricing formats
  • Falls back to model_id when id is missing
• defaultResponseExtractor — handles { data: [...] }, { models: [...] }, and top-level array responses
• formatPrice — formats per-token prices into per-million-tokens dollar strings with adaptive precision
• formatContextLength — formats token counts into compact strings (e.g. 128k, 1M, 1.5M)
• Separate open-model-selector/utils entry point — pure utility functions without "use client" directive, safe for React Server Components
  • Normalizers: defaultModelNormalizer, defaultResponseExtractor, per-type normalizers (normalizeTextModel, normalizeImageModel, …)
  • Formatters: formatPrice, formatContextLength, formatFlatPrice, formatAudioPrice, formatDuration, formatResolutions, formatAspectRatios
  • Helpers: isDeprecated
  • All model types re-exported for normalizer consumers
• Scoped CSS — all variables use --oms- prefix; never pollutes :root or host app styles
  • Light theme defaults
  • Dark mode via @media (prefers-color-scheme: dark) and .dark class ancestor
  • Portal-safe dark mode for popover and tooltip content
• Model tooltip — hover over a model name to see description, context length, provider badge, and input/output pricing
• Comprehensive test suite — unit, component (jsdom), and Storybook (Playwright) test projects
• Storybook stories — Default, PreselectedModel, SystemDefault, CustomPlaceholder, SortByNewest, ControlledFavorites, EmptyState, LoadingState, ErrorState, PopoverTop, WideContainer, DarkMode, MinimalModels, and VeniceLive
• GitHub Actions CI — automated npm publish on GitHub release creation
• TypeScript — full type exports (33 types):
  • Base: ModelType, BaseModel, Deprecation, AnyModel
  • Text: TextModel, TextPricing, TextCapabilities, TextConstraints
  • Image: ImageModel, ImagePricing, ImageConstraints
  • Video: VideoModel, VideoConstraints
  • Inpaint: InpaintModel, InpaintPricing, InpaintConstraints
  • Embedding: EmbeddingModel, EmbeddingPricing
  • TTS: TtsModel, TtsPricing
  • ASR: AsrModel, AsrPricing
  • Upscale: UpscaleModel, UpscalePricing
  • Component / Hook: ModelSelectorProps, TextModelSelectorProps, ImageModelSelectorProps, VideoModelSelectorProps, UseModelsProps, UseModelsResult, FetchFn
  • Normalizer: ModelNormalizer, ResponseExtractor
• Dual CJS/ESM output — built with tsup, sourcemaps and .d.ts included