ai_core_codespark 0.2.0

Local AI model for Flutter — on-device text embeddings & vector search, fully offline, no API keys or cloud. The engine behind semantic search & RAG.

0.2.0

• Typed exceptions — ModelDownloadOfflineException, InsufficientStorageException, and EngineNotInitializedException (all extend CodesparkException), thrown at the right points so you can write precise try/catch around initialize() and embed().
• AI-agent-friendly docs — added "AI Agent Context" guidance to the primary methods (initialize, embed) and an llms.txt describing intent + minimal syntax for LLM tooling.
• Example — added example/lib/download_progress_ui.dart showing a real first-run download UI with progress + error/retry handling.

0.1.1

• Docs: removed decorative emoji from the README for a cleaner, plain-text read.
• Docs: added a "More from ksaikiran.dev" section linking related text, search, and input packages.

0.1.0

Initial release — the on-device embedding foundation for the codespark family.

• CodesparkEngine: one-call API to download, verify, load and run an embedding model entirely on-device (offline, no API keys).
• ModelManager: streamed model download with SHA-256 verification and an atomic, cached, resumable-safe install. Ships a pinned, verified all-MiniLM-L6-v2 (int8, 384-dim) in ModelCatalog.
• BertTokenizer: pure-Dart WordPiece tokenizer, verified byte-for-byte against the HuggingFace reference tokenizer (accents, CJK, Cyrillic, punctuation).
• Embedder + EmbedderIsolate: ONNX inference with mean/cls/max pooling and L2 normalization, run on a background isolate to keep the UI smooth.
• Similarity: cosine, dot product, top-k and MMR (diversity-aware ranking).
• VectorStore: in-memory brute-force search with JSON persistence.