Current API Reference¶
Top-level package¶
mandol.__all__ currently exports:
MemoryUnitMemorySpaceMemorySpaceRegistryTowerSpaceSemanticMapSemanticGraphcluster_nodesClusterMethod__version__
Core helpers¶
mandol.core also provides:
create_semantic_map(text_model=None, image_model=None, embedding_dim=None, preset=None, **kwargs)create_semantic_graph(semantic_map_instance=None, preset=None, **kwargs)create_memory_unit(uid, content, content_type="text", metadata=None, **kwargs)create_memory_space(name, faiss_index_type=None)get_default_core_config()get_model_preset(preset_name)get_core_component_status()validate_embedding_dim(embedding_dim)get_recommended_faiss_index(num_vectors, embedding_dim)
Retrieval package¶
mandol.retrieval exports:
BaseRetrieverRetrievalInterfaceMultiRetrievalInterfaceRetrievalMethodRetrievalResultparse_retrieval_methodsparse_weightsMultiRetrieverBM25RetrieverSPLADERetrieverCosineRetrieverAdapterGraphContextPathInfoGraphContextExpanderRerankerManagerQueryBundleScoreFusion
Three-tower retrieval¶
mandol.triple_retrieval exports:
TripleTowerRetrieverTripleTowerConfigTripleTowerSearchModeRerankStrategyTripleTowerResultHierarchicalTowerResultGraphTowerResultEpisodicTowerResultSecondStageRerankResultTowerRetrievalStatscreate_triple_tower_retriever
Other retrieval-facing packages¶
mandol.hierarchical exposes hierarchical retrieval types and clustering
helpers for already-built hierarchical memory spaces.
mandol.entity_relation exposes EntityRelationGraphRetriever,
GraphRetrievalConfig and create_graph_retriever.
mandol.episodic exposes EpisodicMemoryRetriever,
EpisodicRetrievalConfig and create_episodic_retriever.
High-level memory construction¶
mandol.auto_builder exposes the maintained high-level construction path:
MemoryOrchestratorandOrchestratorConfigHighLevelMemoryBuilderandHighLevelMemoryBuildConfigbuild_high_level_memoryHierarchicalAutoBuilder,EntityRelationAutoBuilderandEpisodicAutoBuilderwith their configuration and result typesPipelineStrategyandSTYLE_STRATEGIES
Routing and quantification¶
mandol.memory_router exposes LocomoTowerRouter,
LongMemEvalTowerRouter and their routing configuration and category types.
These policies are used by the paper’s router + quantification workflows.
mandol.quantification exposes SemanticQuantifier, QueryExpander,
confidence-aware pruning, cascade pruning and the associated factory functions.
Providers, storage and clustering¶
mandol.llm provides LLMClient, LocalLLMClient, provider/model
registries and client factories. mandol.storage provides
RocksDBPayloadStore and TieredStorageManager. mandol.cluster
provides the Leiden and DBSCAN clusterer interfaces used by the builders.