MemPalace/mempalace
Mempalace
Mempalace is a long-term memory management system for large language models that orchestrates the storage and retrieval of conversation history and entity relationships. It functions as a memory orchestrator and Model Context Protocol server, providing AI clients with read and write access to structured knowledge.
The system utilizes a temporal knowledge graph to track evolving entity relationships and timelines with validity windows. It employs a hierarchical memory partitioning strategy, organizing data into wings and rooms to isolate specialist agent contexts and restrict semantic searches to specific topical scopes.
Capabilities include a semantic search vector store for retrieving verbatim conversation history without relying on summaries, as well as pluggable storage backends for using local or external databases. The framework also provides tools for conversation log mining and hook-triggered snapshot backups to prevent data loss during context compression.
Features
- Agent Memory Orchestration - Orchestrates long-term memory and context isolation for specialist agents to prevent system prompt bloat.
- Conversational Memory Systems - Provides a complete system for storing and retrieving long-term user context and interaction history across multiple AI sessions.
- Long-term Memory Stores - Maintains historical state and context through persistent storage to ensure continuity across long-term interactions.
- Model Context Protocol - Implements the Model Context Protocol to allow AI clients to read and write structured knowledge and memory.
- Model Context Protocol Clients - Provides a Model Context Protocol server interface for external clients to execute memory operations.
- Model Context Protocol Servers - Implements a server based on the Model Context Protocol to grant AI clients read and write access to structured knowledge.
- Temporal Knowledge Graphs - Implements a temporal knowledge graph to track evolving entity relationships with validity windows and timelines.
- AI Memory Layers - Manages the architectural layer for storing, backing up, and swapping AI context backends.
- Memory Hierarchies - Uses a hierarchical structure of wings and rooms to restrict memory search queries to specific topics.
- Conversation History Retrieval - Retrieves original, non-summarized conversation history using semantic search for high-precision recall.
- Verbatim Transcript Indexing - Stores original conversation text without summarization to allow precise retrieval via semantic search.
- Semantic Search - Utilizes a semantic search layer with vector embeddings to retrieve verbatim conversation excerpts without relying on summaries.
- Storage Backend Adapters - Supports pluggable retrieval layers to swap between local vector stores and external database services.
- Memory Partitioning Hierarchies - Organizes memory into a structured hierarchy of wings and rooms to restrict semantic searches to specific topics.
- Memory Isolation - Provides logical memory isolation by assigning dedicated data silos to individual specialist agents.
