smart_suggestions_codespark 0.2.1

On-device "related items" for Flutter — suggest similar content by meaning, fully offline, no API keys. Three-line API powered by local embeddings.

0.2.1

• Docs: Added /// comments on all CodesparkBackend members.
• Formatting: dart format on all library files.
• Screenshots: Replaced corrupted free_text_demo.png with a valid PNG.

0.2.0

• SmartSuggestions.create() — one-liner static factory combining constructor + initialize().
• SmartSuggestions.createWithEmbedder() — same pattern for test/custom backends.
• Similarity, Pooling, PoolingStrategy, ScoredIndex now publicly exported for advanced use (custom vector ranking, MMR, normalization).
• Typed exceptions — SuggestionException, IndexOutOfRangeException, IndexDimensionMismatchException replace generic StateError/RangeError.
• Web compatibility — dart:io abstracted behind a conditional import (library compiles on web; persistence throws UnsupportedError on web).
• CI + benchmarks — GitHub Actions workflow (flutter analyze + flutter test) and a benchmark test suite measuring ranking throughput.

0.1.0

Initial release — on-device "related items" suggestions for Flutter.

• SmartSuggestions: one-call API to rank candidates by similarity to an anchor, fully on-device (no API keys, no cloud).
• suggest() / suggestFor<T>(): one-shot suggestions with topK and optional cosine threshold.
• suggestLike(): multi-anchor blending — average multiple reference embeddings (e.g. user history) to suggest related items.
• createIndex() → SuggestionIndex: embed a corpus once and suggest repeatedly; supports similarTo() (within-index, self-excluded), suggestDiverse() (MMR), and incremental add().
• Persistence: SuggestionIndex.save() / SmartSuggestions.loadIndex() — embed once, restore instantly on next launch.
• SuggestionsList<T>: drop-in widget that shows suggestions for a given anchor text or index position.
• Pluggable SuggestionEmbedder (default: CodesparkBackend over ai_core_codespark) — swap in a fake for model-free testing.