Expand description
Vector-based semantic memory with hybrid search and query expansion.
Provides persistent vector storage, local embedding generation, BM25 keyword scoring, hybrid (embedding + BM25) search, and rule-based query expansion for improved recall.
§Main types
VectorStore— Trait for storing and querying embedding vectors.FileVectorStore— File-backed persistent vector store.LocalEmbedding— Local TF-IDF-based embedding provider.HybridSearcher— Combines embedding similarity with BM25 keyword scoring.Bm25Index— BM25 inverted index for keyword-based retrieval.QueryExpander— Trait for expanding queries to improve search recall.RagPipeline— Retrieval-Augmented Generation pipeline for knowledge base search.
Re-exports§
pub use bm25::Bm25Index;pub use conversation::ConversationMemory;pub use conversation::ConversationSummarizer;pub use conversation::ConversationTurn;pub use conversation::CustomerProfile;pub use embedding::EmbeddingProvider;pub use embedding::LocalEmbedding;pub use embeddings_providers::parse_cohere_embedding_response;pub use embeddings_providers::parse_openai_embedding_response;pub use embeddings_providers::parse_voyage_embedding_response;pub use embeddings_providers::BatchEmbeddingProvider;pub use embeddings_providers::CacheStats;pub use embeddings_providers::CachedEmbeddingProvider;pub use embeddings_providers::CohereEmbedV4Provider;pub use embeddings_providers::CohereEmbeddingProvider;pub use embeddings_providers::EmbeddingConfig;pub use embeddings_providers::EmbeddingProviderFactory;pub use embeddings_providers::JinaEmbeddingProvider;pub use embeddings_providers::MistralEmbedProvider;pub use embeddings_providers::NomicEmbedProvider;pub use embeddings_providers::OpenAiEmbeddingProvider;pub use embeddings_providers::SentenceTransformersProvider;pub use embeddings_providers::TogetherEmbedProvider;pub use embeddings_providers::VoyageEmbeddingProvider;pub use hybrid::HybridSearcher;pub use knowledge_graph::Entity;pub use knowledge_graph::EntityType;pub use knowledge_graph::GraphSummary;pub use knowledge_graph::KnowledgeGraph;pub use knowledge_graph::RelationType;pub use knowledge_graph::Relationship;pub use pgvector::PgVectorStore;pub use pinecone::PineconeStore;pub use qdrant::QdrantStore;pub use query_expansion::QueryExpander;pub use query_expansion::RuleBasedExpander;pub use rag::ChunkingStrategy;pub use rag::Document;pub use rag::DocumentChunk;pub use rag::RagConfig;pub use rag::RagPipeline;pub use rag::RagResult;pub use rag::ScoredChunk;pub use store::FileVectorStore;pub use store::InMemoryVectorStore;pub use store::MemoryEntry;pub use store::SearchResult;pub use store::VectorStore;pub use tiered::MemoryContext;pub use tiered::ScoredMemory;pub use tiered::TieredMemory;pub use tiered::TieredMemoryConfig;pub use tiered::TieredTurn;pub use weaviate::WeaviateStore;
Modules§
- bm25
- BM25 inverted index for keyword-based retrieval.
- conversation
- Conversation memory for cross-session customer context. Conversation memory system for cross-session customer context.
- embedding
- Embedding provider trait and local implementation.
- embeddings_
providers - Multiple embedding provider backends (OpenAI, Cohere, Voyage, cached, batch, factory).
Multiple embedding provider backends implementing
EmbeddingProvider. - hybrid
- Hybrid search combining embeddings and BM25.
- knowledge_
graph - Knowledge graph for entity-relationship-based memory. In-memory knowledge graph for entity-relationship storage.
- pgvector
- pgvector (PostgreSQL extension) vector store adapter. pgvector (PostgreSQL extension) vector store adapter.
- pinecone
- Pinecone vector store adapter. Pinecone vector store adapter.
- qdrant
- Qdrant vector store adapter. Qdrant vector store adapter.
- query_
expansion - Query expansion for improved recall.
- rag
- Retrieval-Augmented Generation pipeline. Retrieval-Augmented Generation (RAG) pipeline for knowledge base search.
- store
- Vector store trait and file-backed implementation.
- tiered
- Multi-tier memory: short-term, long-term (episodic), and entity memory. Multi-tier memory system: short-term (working), long-term (episodic), and entity memory.
- weaviate
- Weaviate vector store adapter. Weaviate vector store adapter.