Expand description
§MenteDB: The Mind Database for AI Agents
MenteDB is a purpose-built database engine for AI agent memory. It’s a cognition preparation engine that pre-digests knowledge for single-pass transformer consumption.
§Core Concepts
- MemoryNode: The atomic unit of knowledge (embeddings, graph, temporal, attributes)
- MemoryEdge: Typed, weighted relationships between memories
- MemoryTier: Cognitive inspired storage hierarchy (working, episodic, semantic, procedural, archival)
- Context Assembly: Token budget aware context building that respects attention patterns
- MQL: Mente Query Language for memory retrieval and manipulation
§Quick Start
use mentedb::prelude::*;
use mentedb::MenteDb;
use std::path::Path;
let mut db = MenteDb::open(Path::new("./my-agent-memory")).unwrap();
// store, recall, relate, forget...
db.close().unwrap();§Feature Highlights
- Seven cognitive features: interference detection, pain signals, phantom tracking, speculative caching, stream monitoring, trajectory tracking, write inference
- HNSW vector index with hybrid search (vector + tags + temporal + salience)
- CSR/CSC knowledge graph with belief propagation
- Token budget aware context assembly with attention curve optimization
- MQL query language with vector, tag, temporal, and graph traversal support
- WAL based crash recovery with LZ4 compressed pages
§Repository
Source code: https://github.com/nambok/mentedb
Re-exports§
pub use mentedb_cognitive as cognitive;pub use mentedb_context as context;pub use mentedb_core as core;pub use mentedb_graph as graph;pub use mentedb_index as index;pub use mentedb_query as query;pub use mentedb_storage as storage;
Modules§
- prelude
- Commonly used types, re-exported for convenience.
Structs§
- MenteDb
- The unified database facade for MenteDB.