Namespace JD.AI.Core.Memory
Classes
- BatchEmbeddingPipeline
Processes documents through a chunking → embedding → storage pipeline. Handles batching to stay within embedding provider rate limits.
- InMemoryVectorStore
In-memory vector store for testing and lightweight use cases. Uses brute-force cosine similarity (O(n) per query).
- MemoryConsolidator
Background consolidation service that periodically reads recent daily logs and updates the long-term
MEMORY.mdfile for each project.
- MemoryEntry
A chunk of content stored in the vector store with its embedding.
- MemoryManager
Unified memory manager that combines embedding + vector store.
- MemorySearchResult
Search result from vector similarity search.
- MemoryService
Supports all classes in the .NET class hierarchy and provides low-level services to derived classes. This is the ultimate base class of all .NET classes; it is the root of the type hierarchy.
- SqliteVectorStore
SQLite-based vector store using manual cosine similarity. For production, consider sqlite-vec extension or a dedicated vector DB.
- TextChunk
A chunk of text with position metadata.
- TextChunker
Splits text into overlapping chunks suitable for embedding. Supports configurable chunk size and overlap to maintain context across chunk boundaries.
Interfaces
- IEmbeddingProvider
Abstraction for text embedding providers (OpenAI, Ollama, Voyage, etc.).
- IMemoryService
Provides access to the JD.AI per-project memory directory at
~/.jdai/memory/{projectId}/. Each project has aMEMORY.mdfile for long-term context and amemory/YYYY/MM/subtree for daily turn logs.
- IVectorStore
Vector store abstraction for semantic memory.