Release v2.8.1 · LnYo-Cly/ai4j

ai4j 2.8.1

RAG capability expansion, new vector stores, starter wiring fixes, harness presets — plus the CLI fat-jar move to GitHub Releases.

New vector stores: ElasticsearchVectorStore (ES _bulk/knn/_delete_by_query, lazy dense_vector index; live-verified on ES 9.1.9) and ChromaVectorStore (v2 REST, where dataset scoping; live-verified on Chroma 0.6.3). InMemoryVectorStore gains persistToFile/loadFromFile (atomic via temp file + move).
New chunker: SemanticTextChunker — percentile breakpoints over adjacent-sentence cosine distances via one batch embedding call (verified with a real embedding model).
RAG capabilities: RagQuery.minScore, concurrent multi-variant retrieval, Bm25Retriever.fromChunks, SentenceTextChunker, embedding wiring for ZHIPU/DOUBAO/DASHSCOPE/JINA/BAICHUAN/MINIMAX, generic getStandardRerankService(host,key,url) hatch, getRedisVectorStore() symmetry.
Starter: all vector-store beans are opt-in (ai.vector.<backend>.enabled); new ai.vector.primary + @Primary vectorStore bean resolves unqualified VectorStore injection and fails fast with candidates when ambiguous; ai.vector.elasticsearch.*/ai.vector.chroma.* properties.
Harness: task-level presets and learned-rule promotion (LEARN loop); evidence binds to current execution by default.
CLI: ai4j-cli-*-jar-with-dependencies.jar now ships as a GitHub Release asset (attached below), not on Maven Central.

Pinecone beans are now opt-in — set ai.vector.pinecone.enabled=true after upgrading.
Lombok is provided/optional (since 2.8.0): consumers upgrading from <=2.7 must declare it explicitly.

StandardRerankService usage parsing falls back to meta.tokens/meta.billed_units for Jina-protocol providers (found via a real SiliconFlow call).
Multi-VectorStore injection ambiguity resolved via the primary bean.